Abstract

   graphic file with name ao2c06170_0007.jpg

   Myelodysplastic syndrome (MDS) is difficult to diagnose and classify
   because it has the potential to evolve into acute myeloid leukemia
   (AML). Raman spectroscopy and orthogonal partial least squares
   discrimination analysis (OPLS-DA) are used to systematically analyze
   peripheral blood serum samples from 33 patients with MDS, 25 patients
   with AML, and 29 control volunteers to gain insight into the
   heterogeneity of serum metabolism in patients with MDS and AML. AML
   patients show unique serum spectral data compared to MDS patients with
   considerably greater peak intensities of collagen (859 and 1345 cm^–1)
   and carbohydrate (920 and 1123 cm^–1) compared to MDS patients.
   Screening and bioinformatics analysis of MDS- and AML-related genes
   based on the Gene Expression Omnibus (GEO) database shows that 1459
   genes are differentially expressed, and the main signaling pathways are
   related to Th17 cell differentiation, pertussis, and cytokine receptor
   interaction. Statistical analysis of serological indexes related to
   glucose and lipid metabolism shows that patients with AML have
   increased serum triglyceride (TG) levels and decreased total protein
   levels. This study provides a spectral basis for the relationship
   between the massive serological data of patients and the typing of MDS
   and AML and provides important information for the rapid and early
   identification of MDS and AML.

Introduction

   Hematopoietic stem cells are the source of the malignant clonal illness
   known as acute leukemia.^[40]1,[41]2 On the basis of cell morphology,
   it may be subdivided into acute myeloid leukemia (AML) and acute
   lymphocytic leukemia, each of which has its own subgroups. This kind of
   leukemia is more prevalent in adults and results from a malfunction in
   the bone marrow’s hematopoietic cell proliferation and
   differentiation.^[42]3−[43]5 AML can be clinically classified into
   primary AML and secondary AML. Secondary AML does not meet the standard
   of leukemia at the time of diagnosis but transforms into leukemia as
   the disease progresses. Secondary AML is more difficult to treat than
   primary AML, and the prognosis is worse. Myelodysplastic syndrome
   (MDS), myeloproliferative neoplasms, lymphoma, paroxysmal nocturnal
   hemoglobinuria (PNH), multiple myeloma, and chronic lymphocytic
   leukemia may transform into secondary AML.^[44]6−[45]8

   MDS is a collection of diverse myeloid clonal illnesses that manifest
   as aberrant differentiation and development of myeloid cells and
   originate in hematopoietic stem cells. Hematopoietic failure, low blood
   cell counts, and an increased chance of developing AML are all symptoms
   of this condition.^[46]9,[47]10 Anemia, bleeding, and infection are
   their classical symptoms, and their pathogenesis is not
   clear.^[48]11,[49]12 Clinically, MDS is divided into four distinct
   subtypes: MDS with single lineage dysplasia (MDS-SLD), MDS with
   multilineage dysplasia (MDS-MLD), MDS with excess blasts (MDS-EB), and
   unclassifiable MDS (MDS-U). To further categorize MDS-EB, we
   distinguish between MDS-EB1 and MDS-EB2. AML may develop from bone
   marrow failure in patients with MDS.^[50]13,[51]14 Traditionally, MDS
   is diagnosed based on blood phase and bone marrow phase examination,
   cytogenetic examination, immunological phenotype analysis, and gene
   analysis. There is no “gold standard” for the diagnosis of MDS, which
   is an “exclusive” diagnosis. Its differential diagnosis with AML
   necessitates detection on multiple platforms.^[52]15,[53]16

   Refractory anemia MDS is easily confused with AML, especially the
   MDS-EB subtype. Therefore, it is crucial to investigate a low-cost
   diagnostic tool that may detect myelodysplastic syndromes and AML at an
   early stage in the diagnostic process. In clinical practice, the
   traditional diagnostic methods of MDS and AML are time-consuming,
   expensive, and invasive. Moreover, while the association between
   serological indexes related to glucose and lipid metabolism and the
   transformation from MDS to AML is not backed by sufficient evidence,
   studying this association will allow the rapid and early identification
   of MDS and AML. In this case, developing a rapid identification method
   of MDS and AML without antibody labeling based on Raman spectroscopy
   will allow detailed analysis of the massive serological examination
   data of MDS and AML patients. It will also improve the diagnostic
   efficiency of diseases, reduce detection costs, and promote further
   development of rapid and accurate diagnostic methods identifying the
   transformation of MDS to AML.

   Raman spectroscopy has been used for the rapid, noninvasive, and
   nonlabel identification of hematologic diseases. In a previous study, a
   bifunctional nano probe based on dopant-driven plasma oxide with
   surface-enhanced Raman scattering (SERS) was used to distinguish single
   acute monocyte leukemia THP-1 cells from peripheral blood monocytes
   with high accuracy. This nano probe could be activated by the
   biological redox reaction of single cells to produce stimuli for a
   complementary colorimetric reaction, thus, the prospect of obtaining
   single-cell-level unified identification accuracy is enhanced, allowing
   for the precise and cost-effective detection of cancer cells in complex
   cell samples.^[54]17 Romy et al. differentiated red blood cells (RBCs)
   from erythroid precursor cells by Raman spectroscopy. These precursor
   cells were infected by a SARS-CoV-2 variant in vitro. The authors found
   that differentiated RBCs have impaired hemoglobin biosynthesis,
   abnormal iron metabolism, high serum ferritin levels, and low serum
   iron and transferrin levels, which explains the impaired oxygen-binding
   ability of RBCs in patients with severe COVID-19.^[55]18 The most
   reliable approach for identifying PNH, an uncommon condition marked by
   RBC hemolysis and venous thrombosis, is flow cytometry. Kaan et al.
   devised a technique for analyzing blood samples from participants with
   and without PNH that combines optical tweezers and Raman spectroscopy
   (Raman tweezers). Training using support vector machine analysis
   resulted in an 81.8% specificity and a 78.3% sensitivity for detecting
   PNH with great accuracy.^[56]19 Gold nanoparticles were used in a
   cell-free and label-free SERS approach developed by Stacy et al. for
   the classification of hematopoietic malignancies. When applied to three
   groups, their linear and quadratic discriminant analyses could only
   discriminate between them with a 69.8 and 71.4% accuracy, respectively.
   Their results demonstrate the feasibility of using nanomaterials in
   translational medicine and pave the way for future research into the
   noninvasive monitoring of disease development.^[57]20 Multidrug
   resistance highly correlates with the poor prognosis of chronic myeloid
   leukemia. In a previous study, laser tweezers Raman spectroscopy (LTRs)
   was used to isolate adriamycin-resistant chronic myeloid leukemia cells
   from their parental human chronic myeloid leukemia cell line (K562).
   This study shows that label-free LTRs analysis combined with
   multivariate statistical analysis can be applied to rapidly evaluate
   the chemical resistance status of K562 cells at the single cell
   level.^[58]21

   The examination of serum from individuals with various forms of MDS and
   AML has been performed; however, Raman spectroscopy has not been widely
   employed. Research on the serological indexes, such as glucose and
   lipid metabolism, related to the transformation of MDS to AML is also
   nascent. Therefore, the significance of the present study lies in its
   ability to differentiate between MDS subtypes, primary AML, and
   secondary AML populations based on a clinical model established by
   combining Raman spectroscopy with multivariate analysis. In addition,
   we also looked for biomarkers of MDS and AML among Raman peaks that
   play an important role in disease categorization. The findings of this
   research provide a basis for optimizing the use of data from clinical
   serological tests for the speedy detection of MDS and AML in their
   earliest stages.

Results

Raman Spectroscopic Analysis of Serum in Patients with MDS, Patients with
AML, and Control Subjects

   To study the Raman spectra of the sera of patients with MDS and AML and
   the control group, 173, 197, and 144 Raman spectra of the sera of the
   control group, MDS group, and AML group were obtained, respectively,
   including 42 spectra of patients with MDS-SLD/MLD, 60 spectra of
   patients with MDS-EB1, and 95 spectra of patients with MDS-EB2. The AML
   group included 114 spectra of patients with primary AML and 30 spectra
   of those with secondary AML. [59]Figure [60]1
   A–C shows the serum Raman spectra of control, MDS, and AML groups, MDS
   subtypes, and AML subtypes in the range of 600–1800 cm^–1,
   respectively. Refer to [61]Table S1 for the peak position of relevant
   serum Raman spectra. [62]Figure [63]1A shows the Raman spectra of
   control, MDS, and AML samples. Pink, yellow, and blue vertical lines in
   the figure represent the peaks related to protein (643, 759, 1003,
   1260, 1603, and 1654 cm^–1), nucleic acid (826 and 1579 cm^–1), and
   lipid (1446 cm^–1), respectively. [64]Figure [65]1B shows the Raman
   spectra of control, MDS-SLD/MLD, MDS-EB1, and MDS-EB2 samples, and
   [66]Figure [67]1C shows the Raman spectra of control, primary AML, and
   secondary AML samples, all showing similar peak shapes. According to
   the above spectral patterns and peak positions, it is difficult to
   identify differences in serum components in the control, MDS, and AML
   groups. It is necessary to combine the classification model established
   by orthogonal partial least squares discrimination analysis (OPLS-DA)
   to further screen peak positions that can effectively identify
   potential biomarkers for MDS and AML.

Figure 1.

   [68]Figure 1
   [69]Open in a new tab

   (A) (bottom to top) Mean serum spectra of control subjects, patients
   with MDS, and patients with AML patients. (B) (bottom to top) Mean
   serum spectra of control subjects and patients with MDS-SLD/MLD,
   MDS-EB1, and MDS-EB2. (C) (bottom to top) Mean serum spectra of control
   subjects and patients with primary AML and secondary AML. (D)
   Permutation, cluster, and receiver operating characteristic (ROC) curve
   plots of OPLS discrimination for control vs MDS vs AML (AUC (control) =
   1, AUC (MDS) = 1, and AUC (AML) = 1), MDS subtypes vs primary AML (AUC
   (MDS-SLD/MLD) = 0.75, AUC (MDS-EB1) = 0.842593, AUC (MDS-EB2) =
   0.805556, and AUC (primary AML) = 1), and MDS subtypes vs secondary AML
   (AUC (MDS-SLD/MLD) = 1, AUC (MDS-EB1) = 1, AUC (MDS-EB2) = 1, and AUC
   (secondary AML) = 1) models. (E) Validation plots of OPLS
   discrimination for control vs MDS vs AML, MDS subtypes vs primary AML,
   and MDS subtypes vs secondary AML models.

Establishment of OPLS-DA Model Based on Raman Spectroscopy to Identify MDS
and AML

   From the Raman spectra of control, MDS, and AML groups, 18
   characteristic spectra (6 in each category) were randomly selected to
   form three groups of data. From the Raman spectra of MDS-SLD/MLD,
   MDS-EB1, MDS-EB2, and primary/secondary AML groups, 24 characteristic
   spectra (6 in each category) were randomly selected to form four groups
   of data. The sample data were analyzed and contrasted using supervised
   OPLS-DA. The results of the permutation study indicated that the
   OPLS-DA model was created and was not overfitted since the intercept of
   Q2 on the Y axis was negative. The Raman spectra of control, MDS, and
   AML serum samples, the Raman spectra of MDS-SLD/MLD, MDS-EB1/2, and
   primary AML serum samples, and the Raman spectra of MDS-SLD/MLD,
   MDS-EB1, MDS-EB2, and secondary AML serum samples could all be
   distinguished using cluster analysis in the OPLS-DA model with 100%
   accuracy. The receiver operating characteristic (ROC) curve indicated
   high accuracy of discriminant analysis ([70]Figure [71]1
   D).

   A verification model was utilized to confirm the efficacy of the
   identification approach based on the identification models of control,
   MDS, and AML developed in accordance with the differences in the
   amounts of serum components. A training set and a prediction set made
   up the validation model. The grouping of the spectrum data was
   identified to validate the model, and the training set contained the
   spectra of 12 patients from the same group and 12 patients from a
   different group. The grouping of the spectrum data for the prediction
   set was left unmarked, and it contained the spectra of an additional
   eight patients from the same group and eight patients from a different
   group. Specifically, nine validation models were used to assess the
   diagnostic model’s sensitivity and specificity: control versus MDS,
   control versus AML, MDS versus AML, MDS-SLD/MLD versus primary AML,
   MDS-EB1 versus main AML, MDS-EB2 versus primary AML, and MDS-SLD/MLD
   versus secondary AML. Classification scores for each spectrum in the
   training and prediction sets were calculated using the Raman spectral
   data and the SIMCA-P software. The program awarded classification
   scores to the two groups in the training set based on how the spectral
   data were grouped and then assigned classification scores to the two
   groups in the prediction set based on how closely their spectral data
   matched the spectral data in the training set. The classification was
   considered correct if the first group in the training set and
   prediction set earned a positive score and the second group earned a
   negative score. The categorization was thought to be incorrect in all
   other respects. The diagnostic model’s sensitivity and specificity are
   shown via a point diagram that concurrently shows samples from the
   training set and the prediction set according to the classification
   scores of the two sets ([72]Figure [73]1
   E). Sensitivity and specificity of the classification and
   identification model were achieved by using a cut-off value of zero for
   the predicted value ([74]Tables S4 and S5). Serum sample data from the
   control group, MDS/AML group, MDS subtypes, main AML, and secondary AML
   were all characterized using the OPLS-DA model. The nine validation
   models have a sensitivity range of 75–100% and a specificity range of
   92–100%. The OPLS-DA model was then evaluated based on its potential
   usefulness in the application.

Screening and Validation of Potential Biomarkers

   OPLS-DA score plot, loading plot, and V + S plot for the four models
   (control vs MDS vs AML, control vs MDS, control vs AML, and MDS vs AML)
   are shown in [75]Figure [76]2
   A. The samples may be efficiently divided into three piles in the score
   plot. We were able to clearly differentiate between the control, the
   MDS, and the AML groups since all three clustered in opposite halves of
   the X axis. This finding demonstrates that OPLS-DA can effectively
   differentiate between serum spectral data from healthy controls,
   patients with MDS, and patients with AML, providing the foundation for
   assessing the material properties of the three groups. The Raman peaks
   that are useful for distinguishing between controls, MDS, and AML were
   first screened using the loading plot. Peak values associated with
   protein, nucleic acid, lipids, collagen, and carbohydrates are shown in
   red, yellow, blue, green, and purple, respectively, in the image.
   Specifically, there was a correlation between the two depictions of the
   relationship of sample component content with the component depicted by
   the peak position on the positive semi axis of the loading plot
   ordinate also being relatively higher in the population along the
   positive semi axis of the score plot abscissa. There was a comparable
   correlative link between the negative halves of the loading plot
   ordinate and the scoring plot abscissa. Identifying the three groups of
   samples required the characteristic peaks of protein (897, 1003, 1260,
   and 1660 cm^–1), nucleic acid (726 cm^–1), cholesterol/carotenoid (957
   cm^–1), collagen (859 and 1345 cm^–1), and carbohydrate (920 and 1123
   cm^–1) with the AML group having higher collagen and carbohydrate
   levels than the control and MDS groups, and lower cholesterol levels.
   The value added by peak location to the classification model is
   represented in the V + S graphic by a combination of the VIP and
   correlation coefficient. Future biomarkers were filtered using the
   figure. A further study looked at the peak locations that best
   differentiated the control, MDS, and AML groups to see whether they may
   serve as biomarkers. The V + S plot organized the Raman peaks from
   highest to lowest VIP value. Potential markers were identified by
   screening peaks with VIP > 1.0 and biological importance.

Figure 2.

   [77]Figure 2
   [78]Open in a new tab

   (A) Score plots with 95% Hotelling’s confidence ellipses, loading
   plots, and V + S plots from OPLS models of control vs MDS vs AML,
   control vs MDS, control vs AML, and MDS vs AML. (B) Statistical
   analysis of potential biomarkers from OPLS models of control vs MDS vs
   AML, control vs MDS, control vs AML, and MDS vs AML. (C) Serum
   biochemical analysis of control, MDS, and AML groups. (D) Volcanic map
   of DEGs between MDS and AML groups. The X axis represents the multiple
   changes (logarithmic scale), while the Y axis shows the P value
   (logarithmic scale). Each symbol represents a different gene, and the
   red/blue of the symbol classifies the up/downregulated genes under
   different standards (P value and multiple change threshold). P < 0.05
   was considered to indicate significance. (E) Heatmap of DEGs between
   MDS and AML groups. (F) Bubble diagram of functional enrichment
   analysis of DEGs between MDS and AML groups. The larger the bubble, the
   more genes enriched in that functional pathway, and the closer the
   color of the bubble to green, the higher the significance. (G)
   Biological process, CC, MF, and KEGG enrichment of DEGs between MDS and
   AML groups. (H) Functions and regulatory signaling pathways of genes
   most likely involved in the discrimination between MDS and AML.

   Based on the control vs MDS vs AML model, control, MDS, and AML samples
   were combined in pairs for OPLS-DA, and OPLS-DA score plot, loading
   plot, and V + S plot of control vs MDS, control vs AML, and MDS vs AML
   models were established ([79]Figure [80]2
   A). In each of the three score plots, half of the samples fell on the
   positive X axis, while the other half fell on the negative Y axis. The
   scatter plot clearly displayed sample grouping, demonstrating
   OPLS-superior DA’s ability to extract differential information from
   spectra. The three models were able to distinguish between the two sets
   of samples, and the developed identification approach could detect
   variations in the metabolic components of serum samples. The loading
   plot of the three models showed that the peak intensities of protein
   (897, 1003, 1260, and 1660 cm^–1), nucleic acid (726 cm^–1), and
   cholesterol/carotenoid (957 cm^–1) in the control group were higher
   than those in the MDS group. The peak intensities of protein (897,
   1003, and 1660 cm^–1), nucleic acid (726 cm^–1), collagen (859 cm^–1),
   and cholesterol/carotenoid (957 cm^–1) in the control group were higher
   than those in the AML group, while the peak intensities of collagen
   (859 and 1345 cm^–1) and carbohydrate (920 and 1123 cm^–1) were lower
   than those in the AML group. The peak intensities of protein (897,
   1003, and 1260 cm^–1), nucleic acid (726 cm^–1), collagen (859 cm^–1),
   and carbohydrate (920 and 1123 cm^–1) in the AML group were higher than
   those in the MDS group, while the peak intensity of
   cholesterol/carotenoid (957 cm^–1) was lower than that in the MDS
   group. To identify possible biomarkers in the control, MDS, and AML
   models, the V + S plot of the three models might produce a list of
   Raman peak locations in the order of VIP values from high to low
   ([81]Figure [82]3A). Peaks having biological importance and VIP > 1.0
   were explored for as possible indicators. Important Raman peak
   positions affecting the sample classification were discovered in the
   four models of control vs MDS vs AML, control vs MDS, control vs AML,
   and MDS vs AML. Relevant parameters, such as VIP (VIP > 1.0),
   correlation coefficient, load, and distance, from the center in the V +
   S diagram were also cautiously being considered. Raman peak locations
   without a discernible change were discarded during the following
   biomarker verification step, which removed them from the biomarker
   range.

Figure 3.

   [83]Figure 3
   [84]Open in a new tab

   (A) Volcanic map of DEGs between control and MDS groups. (B) Heatmap of
   DEGs between control and MDS groups. (C) Bubble diagram of functional
   enrichment analysis of DEGs between control and MDS groups. (D) BP, CC,
   MF, and KEGG enrichment of DEGs between control and MDS groups. (E)
   Functions and regulatory signaling pathways of genes most likely
   involved in the discrimination between control and MDS. (F) Volcanic
   map of DEGs between control and AML groups. (G) Heatmap of DEGs between
   control and AML groups. (H) Bubble diagram of functional enrichment
   analysis of DEGs. (I) BP, CC, MF, and KEGG enrichment of DEGs between
   control and AML groups. (J) Functions and regulatory signaling pathways
   of genes most likely involved in the discrimination between control and
   AML.

   The statistical analysis of the peak locations for the Raman
   characteristics with VIP > 1.0 for control vs MDS, control vs MDS vs
   AML, control vs AML, and MDS vs AML is shown in [85]Figure [86]2
   B. [87]Figure [88]2C shows the six types of peripheral blood
   biochemical indexes of total protein (TP), glucose, triglyceride (TG),
   total cholesterol (TC), high-density lipoprotein (HDL), and low-density
   lipoprotein (LDL) in control, MDS, and AML groups. The peak intensities
   of representative proteins (897, 1003, 1206, and 1660 cm^–1), nucleic
   acid (726 cm^–1), collagen (1345 cm^–1), and cholesterol/carotenoid
   (957 cm^–1) in the control group were significantly higher than those
   in the MDS group. The peak intensities of cholesterol/carotenoid (957
   cm^–1) and nucleic acid (726 cm^–1) in the control group were
   significantly higher than those in the AML group, while the peak
   intensities of collagen (1345 cm^–1) and carbohydrate (920 and 1123
   cm^–1) were significantly lower than those in the AML group. The peak
   intensities of collagen (859 and 1345 cm^–1) and carbohydrate (920 and
   1123 cm^–1) in the MDS group were significantly lower than those in the
   AML group. The aforementioned findings supported the serological
   findings for TP, hyperglycemia, and TC, which showed substantial
   variations across groups ([89]Figure [90]2C).

   Bioinformatics was utilized to search the differentially expressed
   genes between MDS and AML, control and MDS, and control and AML. Based
   on Raman spectroscopy and multiparameter analysis, the differentially
   expressed genes’ (DEGs) biological roles and regulatory pathways were
   investigated. DEGs were screened from the [91]GSE15061 chip data based
   on certain screening conditions. There were 456, 26, and 418
   upregulated and 1003, 107, and 981 downregulated DEGs in the
   comparisons of MDS and AML, control and MDS, and control and AML,
   respectively ([92]Tables S6–S11). The red or blue spots on the volcano
   figure ([93]Figures [94]3
   D and [95]4A,F) reflect strongly elevated or downregulated genes,
   respectively. The respective heat maps are shown in [96]Figures [97]2E
   and [98]3B,G. Functional analysis of MDS and AML showed that these DEGs
   were associated with Th17 cell differentiation, pertussis, and cytokine
   receptor interaction pathways.

Figure 4.

   [99]Figure 4
   [100]Open in a new tab

   (A) Score plots with 95% Hotelling’s confidence ellipses, loading
   plots, and V + S plots from OPLS models of MDS subtypes vs primary AML,
   MDS-SLD/MLD vs primary AML, MDS-EB1 vs primary AML, and MDS-EB2 vs
   primary AML. (B) Statistical analysis of potential biomarkers from OPLS
   models of MDS subtypes vs primary AML, MDS-SLD/MLD vs primary AML,
   MDS-EB1 vs primary AML, and MDS-EB2 vs primary AML. (C) Serum
   biochemical analysis of MDS-SLD/MLD, MDS-EB1, MDS-EB2, and primary AML.
   (D) Score plots with 95% Hotelling’s confidence ellipses, loading
   plots, and V + S plots from OPLS models of MDS subtypes vs secondary
   AML, MDS-SLD/MLD vs secondary AML, MDS-EB1 vs secondary AML, and
   MDS-EB2 vs secondary AML. (E) Statistical analysis of potential
   biomarkers from OPLS models of MDS subtypes vs secondary AML,
   MDS-SLD/MLD vs secondary AML, MDS-EB1 vs secondary AML, and MDS-EB2 vs
   secondary AML. (F) Serum biochemical analysis of MDS-SLD/MLD, MDS-EB1,
   MDS-EB2, and secondary AML.

   DAVID was used for gene ontology (GO) and pathway enrichment analysis
   of common DEGs. The DEGs of MDS and AML were significantly enriched in
   three Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Th17
   cell differentiation, pertussis, and cytokine receptor interaction),
   one GO-BP (mitotic spindle organization), three GO-CC (kinetochore,
   condensed chromosome kinetochore, and kinesin complex), and two GO-MF
   (microtubule binding and microtubule motor activity). The top five GO
   functions were obtained after arranging the P values from small to
   large ([101]Figure [102]2
   F,G and [103]Table S12).

   DEGs in the control and MDS groups were significantly enriched in five
   KEGG pathways (hematopoietic cell lineage, T cell receiver signaling
   pathway, Th1 and Th2 cell differentiation, PD-L1 expression, and PD-1
   checkpoint pathway in cancer), four GO-BP (T cell receiver signaling
   pathway, T cell activation, cell surface receiver signaling pathway,
   and positive thymic T cell selection), two GO-CC (alpha–beta T cell
   receptor complexα-βT and T cell receptor complex), and five GO-MF
   (transmembrane signaling receiver activity, T cell receiver binding,
   nonmembrane spanning protein tyrosine kinase activity, protein tyrosine
   kinase activity, and transmembrane receiver protein tyrosine kinase
   activity). The top five GO functions were screened after arranging the
   P values from small to large ([104]Figure [105]3
   C,D and [106]Table S13).

   DEGs of the control and AML groups were significantly enriched in two
   KEGG pathways (T cell receiver signaling pathway and Th17 cell
   differentiation), no GO-BP, three GO-CC (kinetochore, chromosome, and
   central region), and one GO-MF (microtubule binding). The top five GO
   functions were screened after arranging the P values from small to
   large ([107]Figure [108]3
   H,I and [109]Table S14).

   When the results of the three groups of differential gene analysis were
   combined, the GO-BP was related to mitotic spindle and T cell receptor
   signal transduction, CC was kinetochore and chromosome, and MF mainly
   involved microtubule. In KEGG analysis, regulatory signals mainly
   involve the T cell signaling pathway, especially the Th17 cell
   differentiation pathway.

   Based on the signal pathway analysis of DEG, the protein–protein
   interaction (PPI) network was used to identify key candidate genes. The
   PPI network of MDS and AML had 1323 nodes and 27,566 interaction pairs.
   If the topological score of a node was high, it was regarded as a key
   node of the network and the degree values of the top 10 genes were
   determined ([110]Figure S6A and Table S15). The hub genes were
   confirmed by using the cytoHubba plugin as CDK1, CCNB1, IL1B, CCNA2,
   ITGAM, AURKB, TOP2A, KIF11, TLR4, and MAD2L1. The data showed that
   there might be a strong interaction between them. When the results of
   the top five GO functions were combined, among the top 10 genes, KIF11,
   AURKB, CCNB1, and MAD2L1 were involved in mitotic spindle organization;
   AURKB and MAD2L1 mainly related to kinetochore; and MAD2L1 mainly
   related to condensed chromosome kinetochore, and KIF11 mainly related
   to kinesin complex. KIF11 plays a major role in the MFs of microtubule
   binding and microtubule motor activity. IL1B is involved in Th17 cell
   differentiation and cytokine receptor interaction signal pathway.
   ITGAM, IL1B, and TLR4 are involved in the pertussis signaling pathway
   ([111]Figure [112]2
   H).

   The PPI network of control and MDS had 83 nodes and 702 interaction
   pairs. If the topological score of a node was high, it was regarded as
   a key node of the network and the degree values of the top 10 genes
   were determined ([113]Figure S6B, Table S16). The hub genes were
   confirmed by using the cytoHubba plugin as CD8a, CD19, IL7R, CD79A,
   CD2, CCR7, LCK, PAX5, CD247, and CD3D. The data showed that there might
   be a strong interaction between them. When the functional results of
   the top five GO genes were combined, among the top 10 genes, CD8A,
   CD247, and CD3D were involved in T cell receiver signaling pathway;
   CD2, CD8A, CD247, IL7R, and CD3D were involved in cell surface receiver
   signaling pathway; CD3D was involved in positive thymic T cell
   selection; CD3D, CD2, CD79A, CD8A, CD19, CCR7, and IL7R mainly related
   to the external side of the plasma membrane; CD247 and CD3D mainly
   related to the alpha–beta T cell receiver complex; CD79A, CD247, and
   CD3D mainly related to the MF of transmembrane signaling receiver
   activity; LCK is mainly involved in the MF of T cell receiver binding;
   LCK mainly relates to the MF of nonmembrane spanning protein tyrosine
   kinase activity, protein tyrosine kinase activity, and transmembrane
   receiver protein tyrosine kinase activity; CD79A, LCK, CD8A, CD19,
   IL7R, and CD3D are involved in the primary immunodeficiency signaling
   pathway; CD2, CD8A, CD19, IL7R, and CD3D are involved in the
   hematopoietic cell lineage signaling pathway; LCK, CD8A, CD247, and
   CD3D are involved in the regulation of T cell receiver signaling
   pathway; LCK, CD247, and CD3D are involved in the regulation of Th1 and
   Th2 cell differentiation signaling pathway; and LCK, CD247, and CD3D
   are involved in the regulation of PD-L1 expression and PD-1 checkpoint
   pathway in cancer ([114]Figure [115]3
   E).

   The PPI network of control and AML had 1251 nodes and 21,942
   interaction pairs. If the topological score of a node was high, it was
   regarded as a key node of the network and the degree values of the top
   10 genes were determined ([116]Figure S6C and Table S17). The hub genes
   were confirmed by using the cytoHubba plugin as ITGAM, CD8A, CDK1, JUN,
   CCNB1, CCNA2, CD44, MMP9, KIF11, and AURKB. The data showed that there
   might be a strong interaction between them. When the results of the top
   five GO functions were combined, among the top 10 genes, AURKB was
   mainly related to kinetochore, chromosome, and central region; KIF11
   was mainly involved in the molecular function of microtubule binding,
   and JUN was involved in T cell receiver signaling pathway and Th17 cell
   differentiation signaling pathway ([117]Figure [118]3
   J).

   Based on the screening of potential biomarkers of MDS and AML, the MDS
   subtypes model was used to analyze differences in the levels of
   components in the serum samples and screen potential classification
   markers of MDS subtypes and primary AML. [119]Figure [120]4
   A shows the OPLS-DA score plot, loading plot, and V + S plot of MDS
   subtypes vs primary AML, MDS-SLD/MLD vs primary AML, MDS-EB1 vs primary
   AML, and MDS-EB2 vs primary AML. In the score plot, the MDS subtypes
   and primary AML samples were clearly grouped. The MDS subtypes group
   was located in the positive half of the X axis, and the primary AML
   group was located in the negative half of the X axis, showing that the
   MDS subtypes group was distinguished from the primary AML group. This
   result shows that OPLS-DA can well distinguish the serum spectral data
   of patients with MDS-SLD/MLD, MDS-EB1, MDS-EB2, and primary AML, which
   provides conditions for analyzing the material characteristics of the
   four groups. The loading plot was used to preliminarily screen the
   Raman peaks that contribute to the MDS subtypes vs the primary AML
   identification model. The red, yellow, blue, green, and purple peak
   numbers in the figure relate to protein, nucleic acid, lipid, collagen,
   and carbohydrate, respectively. The characteristic peaks of protein
   (853, 1003, 1206, and 1616 cm^–1), lipid (1437, 1443, and 1446 cm^–1),
   nucleic acid (781, 786, and 1485 cm^–1), and carbohydrate (920 cm^–1)
   played important roles in the identification of five groups of samples;
   the main AML group had greater quantities of nucleic acid and
   carbohydrates than the MDS subtypes group. Potential biomarkers were
   further screened using the V + S plot, and the significance of peak
   locations was then analyzed to determine the peak positions that can
   effectively identify potential biomarkers for MDS subtypes and primary
   AML ([121]Figure [122]4A).

   Based on the MDS subtypes vs primary AML model, samples of the primary
   AML and three groups of MDS subtypes were combined for OPLS-DA and the
   OPLS-DA score plot, loading plot, and V + S plot of MDS-SLD/MLD vs
   primary AML, MDS-EB1 vs primary AML, and MDS-EB2 vs primary AML were
   established ([123]Figure [124]4
   A). The two groups of samples in the three score plots were located on
   the positive and negative half of the X axis, respectively. Sample
   clustering in the scatter plot was obvious, indicating that the three
   models could well identify the two groups of samples in the model. As
   seen in the loading plot, the peak intensities of nucleic acid (781,
   786, and 1485 cm^–1) and carbohydrate (920 cm^–1) in the primary AML
   group were generally higher than those in the MDS subtypes group and
   the peak intensities of lipid (1437, 1443, and 1446 cm^–1) were higher
   than those in the MDS-SLD/MLD group and lower than those in the MDS-EB1
   and MDS-EB2 groups. The V + S plot provides the main basis for
   determining the potential classification markers in MDS subtypes and
   primary AML models ([125]Figure [126]4A). Peaks with biological
   importance were chosen as the peak range of putative biomarkers from
   the list of peaks with VIP > 1.0 produced from the V + S plot. In MDS
   subtypes vs primary AML, MDS-SLD/MLD vs primary AML, MDS-EB1 vs primary
   AML, and MDS-EB2 vs primary AML models, the same method was used to
   screen and verify biomarkers.

   [127]Figure [128]4
   B shows the statistical analysis of characteristic Raman peaks with VIP
   > 1.0 of MDS subtypes vs primary AML, MDS-SLD/MLD vs primary AML,
   MDS-EB1 vs primary AML, and MDS-EB2 vs primary AML models. [129]Figure
   [130]4C shows six types of peripheral blood biochemical indexes of TP,
   glucose, TG, TC, HDL, and LDL in the four groups of MDS-SLD/MLD,
   MDS-EB1, MDS-EB2, and primary AML. Statistical analysis showed that the
   peak intensity of representative protein (853, 1003, 1206, and 1616
   cm^–1) in the MDS-SLD/MLD group was significantly higher than that in
   the MDS-EB1, MDS-EB2, and primary AML groups, while the peak intensity
   of representative lipid (1437, 1443, and 1446 cm^–1) was significantly
   lower than that in the MDS-EB1, MDS-EB2, and primary AML groups. These
   results were consistent with those obtained for the serological
   indexes, such as TG, TC, HDL, and LDL, with significant differences
   between groups ([131]Figure [132]4C). The peak intensities of
   representative nucleic acid (781, 786, and 1485 cm^–1) and carbohydrate
   (920 cm^–1) in the MDS-EB1 and MDS-EB2 groups were significantly lower
   than those in the primary AML group.

   Based on the results of screening for potential classification markers
   of MDS subtypes and primary AML, the MDS subtypes model was used to
   screen the potential classification markers of MDS subtypes and
   secondary AML. [133]Figure [134]4
   D shows the OPLS-DA score plot, loading plot, and V + S plot. The four
   groups of samples in the score plot were clearly grouped and
   distributed in the four quadrants of the graph, reflecting that the MDS
   subtypes and secondary AML groups have been distinguished. This result
   shows that OPLS-DA can well distinguish the serum spectral data of
   patients with MDS-SLD/MLD, MDS-EB1, MDS-EB2, and secondary AML, which
   provides conditions for analyzing the material characteristics of the
   four groups. Raman peaks contributing to the MDS subtypes versus the
   secondary AML detection model were first screened using the loading
   plot. Protein, nucleic acid, lipid, collagen, and carbohydrates are
   represented by the red, yellow, blue, green, and purple peak numbers,
   respectively, in the image. The lipid content of the secondary AML
   group was larger than that of the MDS subtypes group, and the protein
   characteristic peaks (1003 and 1616 cm^–1) and lipid peaks (1437 and
   1443 cm^–1) played major roles in the identification of the four groups
   of samples. Further screening of prospective biomarkers was performed
   using the V + S plot, and the importance of typical peak positions was
   verified in future analysis to discover the peak positions that may
   effectively identify potential biomarkers for the control group and MDS
   subtypes ([135]Figure [136]4D).

   Based on the MDS subtypes vs secondary AML model, secondary AML and
   three groups of MDS subtypes samples were combined for OPLS-DA and
   OPLS-DA score plot, loading plot, and V + S plot ([137]Figure [138]4
   D) of MDS-SLD/MLD vs secondary AML, MDS-EB1 vs secondary AML, and
   MDS-EB2 vs secondary AML models were established. The two groups of
   samples in the three score plots were located on the positive and
   negative half of the X axis, respectively, and the sample clustering
   was obvious in the scatter plot. Therefore, it may be considered that
   the two models had good discrimination ability for the two groups of
   samples in the model. It is seen from the loading plot that the
   intensity of the peak position (1003 cm^–1) representing protein in the
   secondary AML group was higher than that in the MDS-EB1 and MDS-EB2
   groups but lower than that in the MDS-SLD/MLD group. The V + S plot
   provides the main basis for determining the potential classification
   markers in the MDS subtypes and secondary AML models ([139]Figure
   [140]4D). According to the V + S plot, a list of peaks with VIP > 1.0
   was derived and peaks with biological significance were selected as the
   peak range of potential biomarkers according to the literature. In MDS
   subtypes vs secondary AML, MDS-SLD/MLD vs secondary AML, MDS-EB1 vs
   secondary AML, and MDS-EB2 vs secondary AML models, the same method was
   used to screen and verify biomarkers.

   [141]Figure [142]4
   E shows the statistical analysis of VIP > 1.0 characteristic Raman peak
   positions of MDS subtypes vs secondary AML, MDS-SLD/MLD vs secondary
   AML, MDS-EB1 vs secondary AML, and MDS-EB2 vs secondary AML.
   [143]Figure [144]4F shows the six types of peripheral blood biochemical
   indexes of TP, glucose, TG, TC, HDL, and LDL in the MDS-SLD/MLD,
   MDS-EB1, MDS-EB2, and secondary AML groups. Statistical analysis showed
   that the peak intensity of representative protein (1003 and 1616 cm^–1)
   in the MDS-SLD/MLD group was significantly higher than that in the
   MDS-EB1, MDS-EB2, and secondary AML groups, while the peak intensity of
   representative lipid (1437 and 1443 cm^–1) was significantly lower than
   that in the MDS-EB1, MDS-EB2, and secondary AML groups. These results
   were consistent with results for the serological indexes, such as TG,
   TC, HDL, and LDL, with significant differences between groups
   ([145]Figure [146]4F).

Discussion

   In this study, OPLS-DA could well distinguish control subjects,
   patients with MDS, and patients with AML based on serum spectral data.
   The results of this exploratory study confirm the feasibility of
   interpreting the heterogeneity of serum metabolism in patients with MDS
   and AML based on serum Raman spectroscopy and multivariate analysis,
   suggesting the utility of serum Raman spectroscopy in the detection of
   MDS, AML, and biomarker mining. The peaks representing protein (853,
   1003, 1206, and 1616 cm^–1), lipid (1437, 1443, and 1446 cm^–1),
   collagen (859 and 1345 cm^–1), and carbohydrate (920 and 1123 cm^–1)
   can be used as potential biomarkers to identify MDS and AML. A decrease
   in peripheral blood cells in patients with MDS due to bone marrow
   failure may cause hypoproteinemia secondary to anemia, putting these
   patients at risk of developing hypoproteinemia. Due to the
   proliferation of malignant cells, the demand for protein and amino
   acids in patients with AML is higher than that in patients in the MDS
   group, and therefore the risk of hypoproteinemia in patients with AML
   is higher.

   High TG levels indicate impaired cellular energy metabolism. Since
   damaged cells cannot efficiently burn sugar, the body must maintain
   high levels of TG in the blood.^[147]22−[148]24 The peripheral serum TG
   level of patients with AML was higher than that in patients with MDS
   and control subjects, suggesting the risk of hyperlipidemia in patients
   with AML. Compared with the MDS-SLD/MLD group, the peak intensity
   representing lipid (1437, 1443, and 1446 cm^–1) in primary AML and
   secondary AML groups was significantly higher, consistent with the
   results of serum biochemical analysis. This may indicate an abnormality
   in cellular energy metabolism in patients with AML, which is consistent
   with our previous findings. According to our previous report on the
   analysis of AML subtypes, the analysis of bone marrow supernatant of
   patients with AML suggested that the presence of leukemia cells in the
   bone marrow microenvironment resulted in lower serum levels of TC, HDL,
   and LDL in patients with AML than in control subjects, indicating that
   AML patients had lipid metabolism disorder.^[149]25

   Statistical analysis revealed that the 957 cm^–1 peak intensity in the
   AML and MDS groups was lower than that in the control group with the
   difference being significant between the AML and the control groups.
   The 957 cm^–1 peak represents both lipid and β-carotenoids. We
   speculate that the antioxidant level in patients with AML and MDS is
   low, which is not conducive to free radical scavenging, while the high
   carotenoid level in healthy individuals inhibits tumors, such as AML
   and MDS.^[150]26−[151]28 In the remission phase of leukemia, the plasma
   carotenoid level increases, which may also act as a self-protection
   mechanism.^[152]29

   Based on the screening of MDS- and AML-related genes in the Gene
   Expression Omnibus (GEO) database, we retrieved ITGAM, IL1B, and TLR4
   genes as key differential genes, which are involved in the pertussis
   signaling pathway. Pertussis is a type of disease that can trigger a
   lymphoid leukemia reaction. Pertussis toxin can inhibit T cell movement
   and displacement by inhibiting the activation of G protein or
   “guanylate binding protein.” It is a special protein related to
   transmembrane signal transduction in the cell membrane. The peak
   intensity of MDS-EB1 in primary AML and MDS subtypes was significantly
   different at 1485 cm^–1. The peak at 1485 cm^–1 is related to guanine
   metabolism. Patients with MDS-EB1 have insufficient T cell function and
   carry a high risk of transforming into AML. Abnormal nucleic acid
   metabolism in the MDS-EB1 metabolic microenvironment may be related to
   the transformation potential of MDS to AML, but further study is
   necessary in this regard.

   Using bioinformatics analysis, we also found that DEGs between MDS and
   AML related to the pathway of Th17 cell differentiation. Th17 is a T
   cell subset that shows heterogeneity and plasticity in different immune
   environments, closely related to the occurrence of autoimmune diseases
   and tumors. The metabolic microenvironment critically influences the
   differentiation and function of Th17.^[153]30 The serine/threonine
   kinase Akt signaling pathway is necessary for the peripheral induction
   of Th17.^[154]31,[155]32 Serine and threonine are hydroxyaliphatic side
   chain amino acids. The difference in the peak position of lipids (1437,
   1443, and 1446 cm^–1) in the blood microenvironment of MDS and AML may
   be related to the transformation from MDS to AML.^[156]33,[157]34

   AML affects the differentiation and proliferation of hematopoietic
   cells due to clonal abnormalities, while MDS affects only the
   differentiation of hematopoietic cells due to clonal abnormalities.
   Therefore, in theory, MDS can be regarded as “pre-leukemia.” In the
   OPLS-DA model, the MDS and AML groups could be well identified and
   cluster analysis yielded good results. The results of Raman spectrum
   analysis of serum samples agreed well with those of serum biochemical
   analysis results, but due to the influence of sample size, serum TP,
   glucose, TC, and LDL levels did not differ significantly, indicating
   that the Raman spectrum detection of serum was more sensitive and could
   capture subtle differences in the component levels that cannot
   otherwise be captured by conventional serological detection.

   The differential diagnosis between MDS and AML is challenging because
   massive serological data related to glucose and lipid metabolism and
   corresponding clinical diagnostic indicators are lacking. The present
   study highlights the need for further mining serological detection
   data. Statistical analysis of larger samples will help to identify new
   methods for the rapid and early identification of different types of
   MDS and AML. Raman spectroscopy, which is both quick and
   cost-effective, will lay a solid foundation for the development of big
   data-aided diagnosis and application software in the future.

   In conclusion, the close combination of serum Raman spectroscopy and
   OPLS-DA can scientifically detect differences in the structure,
   components, and content of biomacromolecules in the serum of patients
   with different types of MDS and AML. Compared with bone marrow
   puncture, Raman spectroscopy analysis of serum samples is less
   traumatic and easily accepted by patients, which is conducive to
   follow-up research. The identification of different types of MDS and
   AML has far-reaching scientific research and clinical significance.

Limitations of the Study

   However, the patient’s medical histories, drug histories, and smoking
   and drinking habits were not thoroughly explored in this research,
   which might have influenced the outcomes. Larger sample size and the
   use of consistent techniques of data collection are therefore required
   to enhance the reliability of samples. And although the control group,
   MDS subgroup, and AML subgroup all had about the same percentage of
   male and female participants, the age distribution was however
   different. The control group had individuals in the age range of 12–69
   years, the MDS-SLD/MLD group had patients who were 19–65 years old, the
   MDS-EB1 group had patients who were 34–77 years old, the MDS-EB2 group
   had patients of age 37–73 years, the primary AML group had patients of
   age 5–64 years, and the secondary AML group had patients who were 34–74
   years old. The difference in age distribution among the six groups was
   statistically significant ([158]Tables S2 and S3). However, the
   metabonomic analysis and statistical methods used to analyze the
   indicators basically met the needs for comparing the differences
   between samples in the study, have prompt value for clinical practice
   guidelines, and also provide a reference for further follow-up
   research.

Conclusions

   To our knowledge, this work is the first application of serum Raman
   spectroscopy to interpret the heterogeneity of serum metabolism in
   patients with MDS and AML. Preliminary analysis of Raman spectra showed
   specific biomolecular differences between MDS and AML, which may be
   caused by changes in the patients’ body metabolism. Peak collagen (859
   and 1345 cm^–1) and carbohydrate (920 and 1123 cm^–1) intensities were
   substantially lower in the MDS group as compared to the AML group.
   Screening and bioinformatics analysis of MDS- and AML-related genes
   based on the GEO database revealed 1459 DEGs, and GO function was
   mainly related to mitotic spindle organization. The main cell
   components were kinetochore, condensed chromosome kinetochore, and
   kinesin complex, and the main molecular function was related to
   microtubule binding and microtubule motor activity. KEGG analysis of
   the DEGs between MDS and AML revealed the main signaling pathways as
   Th17 cell differentiation, pertussis, and cytokine receptor
   interaction. The 10 hub genes screened in the PPI network were CDK1,
   CCNB1, IL1B, CCNA2, ITGAM, AURKB, TOP2A, KIF11, TLR4, and MAD2L1.

   Furthermore, the peak intensity of representative proteins (853, 1003,
   1206, and 1616 cm^–1) in the MDS-SLD/MLD group was significantly higher
   than that in the MDS-EB1, MDS-EB2, and primary AML groups, while the
   peak intensity of representative lipid (1437, 1443, and 1446 cm^–1) was
   significantly lower than that in the MDS-EB1, MDS-EB2, and primary AML
   groups. The peak intensities of representative nucleic acid (781, 786,
   and 1485 cm^–1) and carbohydrate (920 cm^–1) in the MDS-EB1 and MDS-EB2
   groups were significantly lower than those in the primary AML group.
   The peak intensity of representative protein (1003 and 1616 cm^–1) in
   the MDS-SLD/MLD group was significantly higher than that in the
   MDS-EB1, MDS-EB2, and secondary AML group, while the peak intensity of
   representative lipid (1437 and 1443 cm^–1) was significantly lower than
   that in the MDS-EB1, MDS-EB2, and secondary AML group. In particular,
   combined with the statistical analysis of serological indexes related
   to glucose and lipid metabolism, the peaks at 920 and 1123 cm^–1
   closely related to glucose can be used as potential biomarkers for the
   identification of MDS; peaks at 853, 1003, 1206, and 1616 cm^–1 closely
   related to TP, TG, TC, HDL, and LDL, and 1437, respectively, can be
   used to identify AML; and peaks at 1443 and 1446 cm^–1 can be used as
   potential biomarkers for the early diagnosis of MDS-SLD/MLD. The
   results of this exploratory study indicate the potential of applying
   Raman spectrum serum analysis as a clinical tool for the noninvasive
   detection and screening of potential biomarkers to identify MDS and
   AML. The potential correlation between the massive serological
   examination information of patients with MDS and AML and the
   classification and prognosis of the disease was determined. However,
   the sample size of this study was small; therefore, the results may not
   be accurate. Considering that patients and samples with different types
   of MDS and AML are rare and the results are interpretable, we will
   conduct a more detailed prospective study in the next step to test the
   feasibility of identifying biomarkers related to MDS and AML.

Materials and Methods

Sample Collection

   The research included 87 participants, including 44 men and 43 females,
   with age ranging from 5 to 77 years, all of whom were recruited in 2021
   from the Hospital of Blood Diseases, Chinese Academy of Medical
   Sciences (Institute of Hematology, Chinese Academy of Medical
   Sciences). There were 33 patients in the MDS group, including seven
   patients with MDS-SLD/MLD, 10 patients with MDS-EB1, and 16 patients
   with MDS-EB2. There were 25 patients with AML, including 20 patients
   with primary AML and five patients with secondary AML. The control
   group comprised 29 healthy individuals.

   All patients in the experimental group underwent examination by blood
   phase, bone marrow phase, cytogenetics, immunological phenotype
   analysis, and gene analysis, and the results were confirmed by
   experienced hematology experts. The primary diagnosis of these cases is
   hematopoietic diseases, namely MDS or AML. The Chinese Academy of
   Medical Sciences’ Hospital for Blood Diseases Ethics Committee gave its
   approval to this work (KT2020016-EC-2). All participating healthy
   individuals and patients provided written informed consent. Serum
   samples were obtained from patients with MDS, patients with AML, and
   individuals from the control group. All subjects underwent routine
   serum biochemical testing at the clinical testing center of the
   Hospital of Blood Diseases, Chinese Academy of Medical Sciences
   (Institute of Hematology, Chinese Academy of Medical Sciences).

General Biochemical Data

   Serum samples obtained from the subjects who fasted for 10 h were used
   for the analysis of TP, TC, glucose, TG, LDL, and HDL in the peripheral
   blood using an automatic biochemical analyzer.

Raman Spectroscopic Analysis of Peripheral Blood Serum

   On a quartz slide, we deposited 5.0 μL of serum for analysis using a
   confocal Raman spectrometer HORIBA Xplora Raman microscope equipped
   with a 785 nm laser, 40 MW of output power, a 40× objective lens, and
   an XYZ three-dimensional specimen platform. Imaging was performed using
   a 40× (0.75 numerical aperture) Nikon lens, approximately 2 × 2 μm spot
   size range to receive laser beam irradiation, the output power of 10
   MW, the single integration time of 250 s, measurement range of 600–1800
   cm^–1, and resolution of 1 cm^–1. From 5 to 10, various spots were
   measured for each group. To have a baseline, we measured the Raman
   spectra of the quartz slide. Labspec6 was used to perform tasks,
   including smoothing, background subtraction, and baseline correction on
   the collected data. Spectra were all calibrated relative to the Raman
   signal at 1450 cm^–1.

Diagnostic Model Establishment Based on Raman Spectral Data Analysis

   Patients with MDS, AML patients, and healthy controls all had their
   blood Raman spectra analyzed using supervised OPLS-DA performed in
   SIMCA 14.1. The OPLS model’s efficacy was measured by the goodness of
   fit metrics R^2 and Q^2. Under the null hypothesis, we randomly changed
   the Y matrix 200 times and resampled the data to see how well the model
   held up. Raman peaks of statistical significance were identified as
   potential biomarkers using a classification model that included cluster
   analysis and V + S analysis. On the basis of the V + S plot’s
   correlation coefficient, loadings, and distance from the center, the
   peak site of a potential biomarker was selected as a VIP > 1.0. The
   collected candidates for biomarkers were subjected to a significance
   test, and those that had a P value lower than 0.05 were deemed to have
   high clinical utility. Processing of pertinent data was done in Origin.
   Statistics were analyzed using IBM SPSS Statistics 20, and graphics
   were created with GraphPad Prism 5. [159]Figure [160]5
   depicts the conceptual framework for this study, which includes the
   collection of serum samples, the detection of Raman spectra,
   multivariate analysis, and the construction of the identification
   model.

Figure 5.

   [161]Figure 5
   [162]Open in a new tab

   Schematic diagram of serum biochemical analysis by Raman spectroscopy.
   The data set was analyzed by multivariate statistical analysis using
   SIMCA 14.1.

Bioinformatics Analysis

   From the GEO database, we obtained and downloaded gene chip
   [163]GSE15061. DEGs were compared and identified between control and
   MDS, control and AML, and MDS and AML using the R program and the
   Bioconductor software package. P 0.01 and |logFC| > 0.6 were used to
   filter the DEGs. Following this, the screened DEGs were subjected to GO
   function annotation and KEGG signal pathway enrichment analysis using
   DAVID6.8. Cytoscape was used to mine the main genes that play an
   important part in the biological processes connected to alterations in
   the control, MDS, and AML groups using the PPI network analysis
   provided using the STRING online analysis tool.

Statistical Analysis

   To analyze the data, we utilized SPSS 26.0. We compared the frequencies
   using the chi-square test (also called Fisher’s exact test). The mean ±
   standard deviation is used to represent data with a normal
   distribution, whereas the median is used to describe data without a
   normal distribution (25th–75th percentile). The comparison of data
   groups with normal distributions was performed using one-way ANOVA. The
   pairwise comparison of homogenous variance groups was performed using
   the least significant difference approach. The pairwise comparison of
   heterogeneous variance groups was conducted using Tamhane’s T2
   approach. The nonparametric Kruskal–Wallis test technique was used to
   compare groups of data with nonconforming normal distributions. A
   statistically significant difference was defined as P < 0.05.

Acknowledgments