Abstract

   Genome-wide transcriptome profiling after gene perturbation is a
   powerful means of elucidating gene functional mechanisms in diverse
   contexts. The comprehensive collection and analysis of the resulting
   transcriptome profiles would help to systematically characterize
   context-dependent gene functional mechanisms and conduct experiments in
   biomedical research. To this end, we collected and curated over 3000
   transcriptome profiles in human and mouse from diverse gene
   perturbation experiments, which involved 1585 different perturbed genes
   (microRNAs, lncRNAs and protein-coding genes) across 1170 different
   cell lines/tissues. For each profile, we identified differential genes
   and their associated functions and pathways, constructed perturbation
   networks, predicted transcription regulation and cancer/drug
   associations, and assessed cooperative perturbed genes. Based on these
   transcriptome analyses, the Gene Perturbation Atlas (GPA) can be used
   to detect (i) novel or cell-specific functions and pathways affected by
   perturbed genes, (ii) protein interactions and regulatory cascades
   affected by perturbed genes, and (iii) perturbed gene-mediated
   cooperative effects. The GPA is a user-friendly database to support the
   rapid searching and exploration of gene perturbations. Particularly, we
   visualized functional effects of perturbed genes from multiple
   perspectives. In summary, the GPA is a valuable resource for
   characterizing gene functions and regulatory mechanisms after
   single-gene perturbations. The GPA is freely accessible at
   [48]http://biocc.hrbmu.edu.cn/GPA/.
     __________________________________________________________________

   Gene perturbations by knockout, RNA interference (RNAi) or
   overexpression have been widely used to elucidate gene functions,
   considerably impacting many areas of biological and medical research
   over the past decade[49]^1,[50]^2. Huge numbers of gene perturbation
   screens have been performed in many model organisms and in humans. In
   general, these screens focus on detecting molecules associated with
   specific biological phenotypes, such as cell morphology, viability,
   migration and growth rates[51]^3. The recent development of
   high-throughput screening techniques further facilitates the
   comprehensive identification of important genes involved in phenotypes
   of interest. However, it is difficult to directly characterize the
   molecular mechanisms of perturbed genes and depict how perturbed genes
   contribute to specific phenotype changes, such as via interactions with
   other key genes or inducing the dysfunction of specific biological
   processes or pathways[52]^4. Notably, many studies have performed
   transcriptome analysis of expression profiles measured on microarrays
   after gene perturbations. For example, Boumahdi et al. uncovered a gene
   network regulated by SOX2 by analyzing the transcriptome profile of
   SOX2 deletion in squamous-cell carcinoma[53]^5. Through analyzing the
   transcriptome profiles of 147 large intergenic non-coding RNA (lincRNA)
   knockdowns, Guttman et al. revealed that lincRNAs mainly regulated
   global gene expression in trans, maintained the pluripotency and
   repressed the differentiation of embryonic stem cells[54]^6. These
   expression profiles reveal global gene expression changes caused by
   perturbed genes and can be used to infer their context-dependent
   biological functions, cellular pathways and regulatory cascades
   (interacting genes or upstream transcription factors). Thus, it is
   valuable to identify changes of the functions, pathways and regulatory
   cascades through gene perturbation, which provide a unique view of the
   molecular mechanisms of perturbed genes.

   Currently, there are many databases serving gene perturbation
   experiments. Some of these databases provide experimentally validated
   perturbation reagents (e.g., siRNAs), perturbed model organisms (e.g.,
   knockout mouse) or experimental protocols, such as DEQOR[55]^7,
   E-RNAi[56]^8, IKMC[57]^9 and ZFIN[58]^10. Others mainly collect
   phenotype images or descriptions of gene perturbations, such as
   GenomeRNAi[59]^11, IMPC[60]^12, MPD[61]^13. To our knowledge, there is
   no specific database designed to store gene expression profiles
   produced by gene perturbations and perform corresponding transcriptome
   analysis, although the transcriptome profiles of gene perturbations are
   being rapidly accumulated. Thus, the development of such a database
   will greatly promote the discovery of gene function and regulatory
   mechanism, facilitating biological and medical research by experimental
   scientists.

   In this study, we collected and analyzed a large number of
   transcriptome profiles of single-gene perturbations, including
   protein-coding genes, microRNAs and long non-coding RNAs (lncRNAs), in
   human and mouse. Integrating these profiles and corresponding
   transcriptome analysis results, we developed a user-friendly database
   called the Gene Perturbation Atlas (GPA) with several web tools to
   support rapid searching, exploration and visualization of the gene
   perturbations. The GPA provides considerable resources, helping
   biologists to systematically characterize context-dependent gene
   functions and regulatory mechanisms and providing references for