Abstract Background Appropriate treatment of pulmonary hypertension (PH) is critically dependent on accurate discrimination between pre‐ and postcapillary PH. However, clinical discrimination is challenging and frequently requires a right heart catheterization. Existing risk scores to detect postcapillary PH have suboptimal discriminatory strength. We have previously shown that platelet‐derived RNA profiles may have diagnostic value for PH detection. Here, we hypothesize that platelet‐derived RNAs can be employed to select unique biomarker panels for the discrimination between pre‐ and postcapillary PH. Methods and Results Blood platelet RNA from whole blood was isolated and sequenced from 50 patients with precapillary PH (with different PH subtypes) as well as 50 patients with postcapillary PH. RNA panels were calculated by ANOVA statistics, and classifications were performed using a support vector machine algorithm, supported by particle swarm optimization. We identified in total 4279 different RNAs in blood platelets from patients with pre‐ and postcapillary PH. A particle swarm optimization–selected RNA panel of 1618 distinctive RNAs with differential levels together with a trained support vector machine algorithm accurately discriminated patients with precapillary PH from patients with postcapillary PH with 100% sensitivity, 60% specificity, 80% accuracy, and 0.95 (95% CI, 0.86–1.00) area under the curve in the independent validation series (n=20). Conclusions This proof‐of‐concept study demonstrates that particle swarm optimization/support vector machine–enhanced classification of platelet RNA panels may be able to discriminate precapillary PH from postcapillary PH. This research provides a foundation for the development of a blood test with a high negative predictive value that would improve early diagnosis of precapillary PH and prevents unnecessary invasive testing in patients with postcapillary PH. Keywords: blood platelet, diagnosis, machine learning, pulmonary hypertension, RNA Subject Categories: Pulmonary Hypertension, Biomarkers, Pulmonary Biology, Platelets __________________________________________________________________ Nonstandard Abbreviations and Acronyms CTEPH chronic thromboembolic pulmonary hypertension mPAP mean pulmonary arterial pressure PAH pulmonary arterial hypertension PAWP pulmonary arterial wedge pressure PH pulmonary hypertension PSO particle swarm optimization RHC right heart catheterization Clinical Perspective. What Is New? * Blood platelet RNAs have differential levels in patients with pre‐ as compared with postcapillary pulmonary hypertension. * A platelet RNA biomarker panel enables identification of patients with pre‐ and postcapillary pulmonary hypertension. What Are the Clinical Implications? * Platelet RNA profiles may serve as a biomarker in the detection of precapillary pulmonary hypertension. * This could be used in the future to decrease the amount of invasive right heart catheterization in the diagnostic workup of patients with pulmonary hypertension, and further prospective validation is required. Pulmonary hypertension (PH) is described as an increase of mean pulmonary arterial pressure ≥20 mm Hg at rest as assessed by right heart catheterization (RHC).[49] ^1 PH is hemodynamically classified into pre‐ and postcapillary subtypes on the basis of a threshold value for pulmonary arterial wedge pressure (PAWP) of 15 mm Hg[50] ^1 (Figure [51]1A). The accurate distinction between pre‐ and post‐capillary PH is crucial because clinical management of the 2 conditions is profoundly different. Patients with precapillary PH, such as those with pulmonary arterial hypertension (PAH) and chronic thromboembolic PH (CTEPH), are known to be treatable with specific interventions aimed at lowering pulmonary vascular resistance. Postcapillary PH is often diagnosed without performing RHC and on the basis of signs of left ventricular dysfunction or left‐sided valvular disease on echocardiography. In patients without apparent symptoms of left‐sided heart disease, distinguishing between pre‐ and postcapillary PH is challenging. Previous research indicated that 12%–19% of patients referred to a tertiary PH center for RHC have an elevated PAWP, indicating postcapillary PH, despite a prior suspicion of precapillary PH.[52] ^2 Hence, easier discrimination between pre‐ and postcapillary PH is desired. Figure 1. Platelet RNA panels enable the ability to distinguish between pre‐ and postcapillary pulmonary hypertension. Figure 1 [53]Open in a new tab A, Schematic overview of two PH hemodynamic classes (created by [54]www.Biorender.com). B, Volcano plot illustrating that 49 RNAs had increased levels (FDR <0.05, logFC >1) (Red dots) and 219 RNAs had statistically significant decreased levels (FDR<0.05, logFC<−1) (blue dots) in the precapillary PH group. C, Heatmap with unsupervised clustering of the platelet mRNA profiles of precapillary PH (red) and postcapillary PH group (blue) included in all the series. D, The bar plot represents the top 10 up‐ and downregulated platelet RNAs. X axis: Log fold change in the range of −4 to 4. The colors represent the FDR. E. Gene enrichment analysis. CTEPH indicates chronic thromboembolic pulmonary hypertension; FDR, false discovery rate; logFC, logarithm fold change; mPAP, mean pulmonary arterial pressure; NTA SMG1, novel transcript antisense to SMG1; PAWP, pulmonary arterial wedge pressure; PH, pulmonary hypertension; PVR, pulmonary vascular resistance; and SRP, signal‐recognition particle. Several scoring models have been developed to facilitate the noninvasive prediction of postcapillary PH. For instance, the H[2]FPEF score, which stands for heart/hypertensive, atrial fibrillation, pulmonary hypertension, elder, and filling pressure, was created to distinguish between heart failure with preserved ejection fraction and noncardiac causes of dyspnea in a predominantly heart failure with preserved ejection fraction population.[55] ^3 This model employed clinical features and echocardiographic measures and correctly recognized 48% of patients with postcapillary PH and 92% with a high probability of precapillary PH. The left heart failure risk score was developed to predict elevated PAWP; however, only 20% of postcapillary PH were detected when using the score.[56] ^4 We recently refined the left heart failure risk score (renamed to OPTICS risk score) on the basis of clinical criteria to detect elevated PAWP in patients with PH without clear signs of left‐sided heart disease. However, only 23% of patients with postcapillary PH were detected, which may show that clinical scores are inadequate in making the correct distinction between both groups.[57] ^5 At present, no blood‐based biomarker accurately discriminates between pre‐ and postcapillary PH.[58] ^6 The plasma protein marker NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide) is the only prognostic factor currently recognized for PH conditions but lacks diagnostic specificity, particularly when pulmonary artery pressures are only mildly elevated (mean pulmonary arterial pressure [mPAP] <35 mm Hg). NT‐proBNP does not discriminate between pre‐ and postcapillary PH.[59] ^6 A recent study indicated that circulating cell‐free DNA increases in patients with PAH, which is relevant to disease severity and predicts worse survival.[60] ^7 Recently, we showed that platelet RNA can function as a biomarker trove to detect and discriminate patients with PH from asymptomatic controls.[61] ^8 Blood platelets are circulating anucleated cell fragments that are classically known for their role in hemostasis and initiation of wound healing. Platelets contain premature messenger RNA and mediators of RNA transcription, splicing, and translation. This indicates that platelets are highly metabolically active even while in circulation and that RNA processing and translation do occur in these cellular fragments. Premature platelet RNA is subjected to cue‐specific splicing events derived from microenvironmental changes, such as those induced by cancer, infection, or inflammatory diseases.[62] ^9 We and others have previously shown that platelets become “educated” by specific RNA splicing events during disease,[63] ^10 which allows detection of the presence of PH.[64] ^8 In this follow‐up study, we hypothesized that platelet RNA signatures might enable accurate blood‐based discrimination between pre‐ and postcapillary PH. Methods The raw sequencing data in FASTQ format has been deposited in the National Center for Biotechnology Information's Gene Expression Omnibus database under accession number GEO: [65]GSE229083. The source code for the thromboSeq algorithms, which encompasses the thromboSeq dry‐lab pipeline and code for reproducing the main figures in the article, can be accessed through this link: [66]https://github.com/MyronBest/thromboSeq_source_code_v1.5. Clinical Data and Study Cohort Selection This study was approved by the institutional review board and medical ethics committee (2015.220) of the Amsterdam University Medical Center, VU University Medical Center, Amsterdam, The Netherlands (a tertiary referral center for PH). Between October 2015 and May 2021, patients visited the Amsterdam University Medical Center for a diagnostic RHC. All patients provided informed consent. Diagnosis of PH and PH classification in all patients followed standard criteria indicated by the current guidelines.[67] ^1 In this study, the classical definition of PH (>25 mm Hg) was used, not >20 mm Hg, because specific treatments are currently proven only when mPAP is >25 mm Hg. Patients included with hemodynamic criteria by considering precapillary PH (with mPAP >25, pulmonary vascular resistance >3 WU, and PAWP <15), isolated postcapillary PH (with mPAP >25, PAWP >15 and pulmonary vascular resistance ≤3 WU) or combined pre‐ and postcapillary PH (mPAP >25 mm Hg, mPAWP >15 and pulmonary vascular resistance >3 WU). The latter 2 groups were combined and are referred to as postcapillary PH. This diagnosis was made if the PAWP was >15 mm Hg at rest or elevated to >18 mm Hg immediately following 500 mL of saline infusion over 5 minutes.[68] ^1 If no reliable wedge was obtained, left ventricular end‐diastolic pressure was measured. Whole venous blood from patients with a diagnosis of precapillary PH (ie, idiopathic PAH, hereditary PAH, CTEPH), and postcapillary PH (ie, left heart disease PH) was collected from the internal jugular vein during RHC according to current guidelines.[69] ^1 Blood samples drawn during RHC were collected after diagnosis in two 4‐mL or one 6‐mL EDTA‐coated purple‐capped BD Vacutainers (catalog no. 367 863, BD). Samples were processed following standard protocols for platelet isolation, using 2‐step centrifugation at room temperature within 24 hours after blood collection, using a previously established and standardized methodology, with minimal leukocyte contamination and platelet activation (Figure [70]S1A).[71] ^11 Following blood collection in EDTA‐coated tubes and the platelet‐isolation process, the absence of platelet activation was previously validated by flow cytometry analysis of P‐selectin and CD63 on the platelet membrane.[72] ^12 Platelet RNA Processing and Bioinformatics Analyses The platelet RNA was isolated, quality checked, and processed toward RNA‐sequencing libraries as extensively described previously and discussed in Data [73]S1. We employed our standardized thromboSeq pipeline for the bioinformatics processing and differential expression analyses, as well as support vector machine algorithm development.[74] ^11 We used the setting for particle swarm‐optimization (PSO)‐enhanced algorithm development previously published.[75] ^11 We selected the particle (algorithm settings) with the best performance in the evaluation series following the evaluation of 100 particles during 10 iterations (1000 particles in total; additional details are provided in Data [76]S1). Results Platelet RNA Profiles Allow for Blood‐Based Discrimination Between Pre‐ and Postcapillary PH We first investigated whether different PH subgroups may differentially educate platelets resulting in different spliced RNA profiles. For this, we collected and isolated platelet pellets from whole blood of 50 patients with idiopathic PAH or hereditary PAH (n=30) and CTEPH (n=20) as well as from 50 patients with postcapillary PH. Patients with postcapillary PH were older and had significantly higher body mass index and more cardiovascular comorbidities. Pulmonary hemodynamics revealed a significantly lower mPAP and a higher pulmonary vascular resistance and mean right atrial pressure in patients with precapillary PH. An overview of the demographic characteristics and clinical parameters is shown in the [77]Table (see also Table [78]S1). Table . General Characteristics and Hemodynamics of the study cohort Patients Pre capillary PH Post capillary PH Statistical (n=50) (n=50) significance Gender, F/M, n (%) 31 (62%)|19 (38%) 37 (74%)|13 (26%) NS Age, mean±SD 55±17 67±10 <0.001 BMI, mean±SD 26.7±6.1 33.5±7.3 <0.001 Medical history Diabetes Mellitus n (%) 7 (14) 16 (32) 0.03 Atrial fibrillation n (%) 2 (4) 22 (44) <0.001 Hypertension n (%) 13 (26) 27 (54) 0.002 Dyslipidemia n (%) 3 (6) 5 (10) NS Valvular surgery without residual left valvular disease n (%) 1 (2) 2 (4) NS Coronary artery disease n (%) 1 (2) 16 (32) <0.001 Left heart disease n (%) 6 (12) 42 (84) <0.001 RHC mRAP, mm Hg median(IQR) 7 (4) 11 (6) 0.0013 mPAP, mm Hg 47±12.5 39±12.4 0.0016 Cardiac output, L/min 5.3±1.6 5.7±1.6 NS Heart rate, beats/min 78±11 74±13 NS PVR, dyn.s.cm^‐5 median(IQR) 515 (423) 250 (195) <0.001 Wedge pressure, mm Hg median (IQR) 11 (6) 19 (5) <0.001 [79]Open in a new tab Data are given as mean±SD, median (IQR), or n (%). Comparing patient characteristics between patients with pre‐ and postcapillary PH was done using Student t‐tests and X^2 tests. BMI indicates body mass index; IQR, interquartile range; mPAP, mean pulmonary arterial pressure; mRAP, mean right atrial pressure; NS, not significant; PH, pulmonary hypertension; PVR, pulmonary vascular resistance; and RHC, right heart catheterization. Next, the platelet samples were subjected to RNA sequencing. The patients with precapillary PH had on average 4105 different RNAs detected, whereas patients with postcapillary PH had 4000 RNAs detected (Figure [80]S2A and S2B). All samples passed our quality‐control measures. In total, 268 RNAs were identified, with differential expression levels among the 2 groups (false discovery rate <0.05, logarithm fold change >1, and logarithm fold change <−1) (see also Table [81]S2). Forty‐nine RNAs had statistically significantly increased levels, and 219 RNAs were decreased in the precapillary PH group (Figure [82]1B). Unsupervised hierarchical clustering of RNAs with differential splice junction reads resulted in a separation between the 2 groups (Figure [83]1C; P<0.0001, Fisher's exact test subjected to PSO optimization). The top 10 up‐ and downregulated platelet RNAs are shown in Figure [84]1D and Table [85]S3. To comprehend potential mechanistic insight into generated RNA panels between patients with precapillary PH and patients with postcapillary PH, we performed pathway enrichment analysis employing the R software package Enricher (R Foundation for Statistical Computing, Vienna, Austria) (Table [86]S4). The top 10 selected pathways indicated that platelets from patients with precapillary PH were enriched for RNAs implicated in processes such as RNA translation and ribosomal RNA processing (P<0.05). Moreover, these analyses indicated that the RNAs enriched in the precapillary PH group are involved in neutrophil degranulation and regulation of immune response (P<0.05), which may be correlated to inflammatory conditions (eg, vascular inflammation) implicated in the pathophysiology of precapillary PH (ie, PAH and CTEPH). Viral transcription pathways are enriched in the precapillary PH group. Previous studies show that PAH pathogenesis could be the result of the interaction of multiple modulating genes with environmental factors. There is a well‐documented association of viruses with angioproliferation and the development of PAH.[87] ^13 Additionally, precapillary PH RNAs were enriched for proteolysis pathways (P =0.002) with a key role in extracellular matrix composition. Inflammation causes an imbalance in proteolytic enzymes and their inhibitors, resulting in extracellular matrix remodeling and endothelial dysfuncion in the pulmonary vasculature.[88] ^14 The presence of dysfunctional von Willebrand factor that causes endothelial disruption in PAH has been linked to increased proteolysis[89] ^15 (Figure [90]1E). Therefore, perhaps endogenous retroviruses may be activated in platelets in the pathophysiological process of precapillary PH. Thus, platelet RNA profiles may discriminate patients with pre‐ and postcapillary and are associated with specific RNA processing and immune and proteolysis pathways. As patients with postcapillary PH are inherently different from those with precapillary PH in terms of age, body mass index, and multiple comorbidities (see [91]Table and Table [92]S1), we cannot rule out contribution of those variables to the biomarker panel. In all, correlative analysis showed only few RNAs to be weakly associated with these clinical variables (Figure [93]S3). Among those of interest are TMSB4X pseudogenes, of which the parent gene TMSB4X was previously implicated in the angiogenesis process[94] ^16 and a circular RNA isoform of TMSB4X as a biomarker for pediatric PAH.[95] ^17 PSO‐Enhanced thromboSeq Algorithm for the Detection of Patients With Pre‐ and Postcapillary PH We next devised to develop an RNA profile–based classifier that separates patients with precapillary PH from postcapillary PH by platelet RNA analysis. Such a classifier may in the future be employed to independently classify new patients with the diagnostic question whether this is pre‐ or post‐capillary PH at hand. For this, the total sample set (N=100) was separated into a training (n=20 pre‐ and n=20 postcapillary PH), evaluation (n=20 pre‐ and n=20 postcapillary PH), and validation series (n=10 pre‐ and n=10 postcapillary PH); see Figure [96]2A. Our previously designed software is based on a support vector machine algorithm, supplemented by a PSO step, the latter optimizing multiple thresholds during the training process, ultimately reaching the most optimal RNA biomarker panel. The algorithm employs training and evaluation series to iteratively search for the most optimal RNA panel separating both conditions (pre‐ and postcapillary PH), after which the machine learning algorithm is locked, and the validation series is validated (Figure [97]2A).[98] ^11 Following algorithm development, the classifier included an RNA biomarker panel of 1196 markers, resulting in an area under the curve (AUC) of 0.91 in the training series (as measured by the Cortes & Mohri method[99] ^18 ; 95% CI, 0.82–1.00; sensitivity, 75%; specificity; 90%; n=40), AUC of 0.90 in the evaluation series (95% CI, 0.80–1.00; sensitivity, 100%; specificity, 70%; n=40), and an AUC of 0.95 in the validation series (95% CI, 0.85–1.00; sensitivity, 100%; specificity, 60%; n=20; Figure [100]2B). The optimal cutoff thromboSeq algorithm score of 0.36 was identified—aiming at maximum correct classification of patients with precapillary PH—resulting in a sensitivity of 100% (95% CI, 0.83–1.00; n=20) and specificity of 70% in the evaluation series (95% CI, 0.45–0.88; n=20). Once this cutoff was applied to the validation series, this resulted in a sensitivity of 100% (95% CI, 0.69–1.00; n=10) and a specificity of 60% for the test (95% CI, 0.26–0.87; n=20; Figure [101]2C through [102]2E). As a control, random sampling of alternative training and evaluation series (n=1000 iterations) resulted in similar classification accuracies (AUC validation series, 0.91; interquartile range, 0.04), whereas assigning random diagnostic group labels to the samples in the training series, expecting nonsense random classifications, resulted in diminished classification accuracies (AUC, 0.55; interquartile range, 0.15; P<0.001). We thus conclude that a highly sensitive rule‐out classification algorithm can be developed to identify patients with precapillary PH. Since the previous noninvasive parameter prediction model (OPTICS score) demonstrated a good distinction between the 2 PH subtypes, we next investigated whether the OPTICS score was able to recognize patients with pre‐ and postcapillary PH in the current validation series. For this, the OPTICS scores were calculated for all included samples in the validation series, using a predefined cutoff of >104.[103] ^5 The precapillary PH group was predicted with 100% sensitivity (n=10) with both the OPTIC prediction model and the thromboSeq algorithm. However, only 30% of patients with postcapillary PH were predicted correctly by the OPTICS model, whereas the thromboSeq algorithm was able to detect 60% of these. Figure 2. PSO‐enhanced thromboSeq algorithm development, optimization, and validation, for discrimination of pre‐ and postcapillary PH. Figure 2 [104]Open in a new tab A, Schematic representation of the samples series used to develop the PSO‐enhanced thromboSeq algorithm. Pre‐ and postcapillary PH samples were divided into 3 different groups: training (gray), evaluation (black), and validation (orange) series. Training and evaluation series were employed for algorithm training and optimization. An independent cohort of samples (validation series) was used to evaluate the performance of the test. B, Receiver operating characteristic (ROC) curves of the thromboSeq algorithm of the training (gray line), evaluation (black line), and validation (orange line) series. Indicated are the sample number per series (n), area under the curve (AUC) values, and the 95% CIs. C, Predicting scores for the samples among the different series. Classification scores closer to 0.0 corresponds to the prediction of postcapillary PH (blue bars) and 1.0 to precapillary PH (red bars). The orange line indicates the algorithm thresholds (0.36) selected using the evolution series setting. D, Cross table of support vector machine/swarm for pre‐ and postcapillary PH in the training, evaluation, and validation series. E, Detection accuracy in the training, evaluation, and validation series. PH indicates pulmonary hypertension; and PSO, particle swarm optimization. Discussion Diagnosing pre‐ and postcapillary PH is hampered by nonspecific clinical signs and symptoms and often inconclusive echocardiography reports. In many patients, an invasive and costly follow‐up diagnostics test such as RHC is undesirable. Additional biomarkers and scoring tools may supplement the diagnostic path and may improve patient selection for further diagnostic workup. Here, we provide proof‐of‐concept that platelet RNA profiles may be employed to distinguish pre‐ and postcapillary subtypes of pulmonary hypertension. Previous studies have demonstrated a role for megakaryocytes and platelets in PH.[105] ^19 Platelet counts are a valuable marker for risk stratification and closely predict survival of patients with idiopathic pulmonary hypertension.[106] ^20 Moreover, there are multiple signs of platelet involvement in patients with PAH and CTEPH, such as low platelet counts and a high mean platelet volume.[107] ^19 In addition, the interaction of platelets with several types of cells such as pulmonary artery endothelial cells and smooth muscle cells implicated in the pathogenesis of PAH has previously been shown to have a role in vascular remodeling, inflammation, and thrombosis.[108] ^21 In this study, we identified enrichment of RNAs in platelets from patients with precapillary PH that are associated with neutrophil degranulation and immune response regulation pathways. This finding may be correlated to inflammatory conditions (eg, vascular inflammation) implicated in the pathophysiology of precapillary PH (ie, PAH and CTEPH). Moreover, proteolysis pathways enriched in precapillary RNAs have critical roles in extracellular matrix composition and can lead to vascular remodeling and stiffness.[109] ^22 Though these gene ontology analyses may provide some biological insight into the underlying mechanism, they should be interpreted with caution since no functional experiments were performed to confirm these hypotheses. Such experiments may be of interest to investigate the potential role of platelets in PH development. This can be done, for example, by measuring associated immune markers in plasma, investigating neutrophil content, profiling the protein content of platelets, perhaps also using murine models of PH with and without the use of splice inhibitors and translation inhibitors. In this study, we included all patients with PH who were subjected to a diagnostic or follow‐up RHC in our center. Such patient selection may introduce bias, as all patients included were suspected of, or were being treated for, PH. Second, we did not discriminate between isolated postcapillary and combined pre‐ and postcapillary PH. We included 23 patients with isolated postcapillary PH and 27 patients with combined pre‐ and postcapillary PH. This relatively low number of samples does not allow a distinction between these 2 subphenotypes. On the other hand, since the PAWP cutoff in the RHC assessment of 15 is rather arbitrary, it could be argued that a strong biological phenotype might outweigh the invasive test. Therefore, adding a biological aspect in the diagnostic process may help to underpin the distinction among 3 hemodynamic PH groups (precapillary, isolated postcapillary, and combined pre‐and postcapillary PH). The sample size of 100 patients in this study is rather small. While many RNA expression‐based machine learning algorithms are trained on hundreds of samples, we could not collect such large numbers within a reasonable time frame. While only a small validation series of 20 samples was assessed in this study, our results suggest that platelet spliced RNA profiles may be able to discriminate pre‐ and post‐capillary PH and may be of additive value to other clinical scores. The cutoff of the biomarker can be adjusted for enhanced sensitivity or specificity performance; however, more data are required to further improve the accuracy of the algorithm. The clinical variables of patients with pre‐ and postcapillary PH were significantly different due to the distinct pathophysiology of the 2 subtypes of PH. The contribution of these variables to the selected biomarker panel cannot be ruled out due to the limited sample size. The correlation analysis between RNA markers and clinical variables revealed only a few RNAs that were positively or negatively correlated with variables such as age, body mass index, comorbidities (atrial fibrillation, left heart disease, and coronary artery disease), and hemodynamic parameters (mPAP and mean right atrial pressure). Of note, due to the small sample size, we cannot rule out the contribution of clinical factors to the biomarker panel, though correlative analysis suggested only rather minor associations between the measured RNAs and clinical variables. Also, all machine learning–based algorithms are at risk of overfitting to natural variation in the current population. To avoid overfitting, we chose a postulated particle based on defined criteria and were blinded to the outcome. We confirmed the stability and generalizability of the classification algorithm by multiple bootstrapping experiments. Despite this, independent validation of the findings in this proof‐of‐concept study in a population with a larger and true medical suspicion of PH is still required before using the platelet biomarker as a diagnostic tool. We are currently expanding our cohort to conduct such a study, preferably in a prospective and multicenter manner. Such studies should ideally also evaluate the combined use of platelet RNA biomarkers with clinical and other diagnostic modalities. Conclusions In conclusion, our study provides proof of concept that by using platelet RNA sequencing and a PSO/support vector machine algorithm, a discrimination can be made between patients with post‐ and precapillary PH (predominantly PAH and CTEPH). We believe that platelet transcriptomes may serve as a complex biomarker in the detection of precapillary PH and may, in the future, be implemented in the diagnostic workup of patients with PH. Such a strategy may reduce the total number of invasive RHC procedures. Validation within a larger prospective cohort and a more thorough understanding of platelet function in PH remain necessary. Sources of Funding This research was supported by the Netherlands Cardio Vascular Research Initiative (CVON‐2012‐08 PHAEDRA, CVON‐2018‐29 PHAEDRA‐IMPACT, CVON‐2017‐10 DOLPHIN‐GENESIS). Financial support was provided by Stichting STOPHersentumoren.nl. The authors acknowledge the receipt of grant support from Ferrer, Janssen, and Merck Sharp & Dohme. The content is solely the responsibility of the authors. Disclosures None. Supporting information Data S1 [110]Click here for additional data file.^ (2MB, pdf) Tables S1–S4 [111]Click here for additional data file.^ (22.2KB, xlsx) Figures S1–S3 [112]Click here for additional data file.^ (47.6KB, xlsx) Acknowledgments