Search

Explainable artificial intelligence aims to interpret how machine learning models make decisions, and many model explainers have been developed in the computer vision field. However, understanding of the applicability of these model explainers to biological data is still lacking. In this study, we comprehensively evaluated multiple explainers by interpreting pre-trained models for predicting tissue types from transcriptomic data and by identifying the top contributing genes from each sample with the greatest impacts on model prediction. To improve the reproducibility and interpretability of results generated by model explainers, we proposed a series of optimization strategies for each explainer on two different model architectures of multilayer perceptron (MLP) and convolutional neural network (CNN). We observed three groups of explainer and model architecture combinations with high reproducibility. Group II, which contains three model explainers on aggregated MLP models, identified top contributing genes in different tissues that exhibited tissue-specific manifestation and were potential cancer biomarkers. In summary, our work provides novel insights and guidance for exploring biological mechanisms using explainable machine learning models.

Massive parallel DNA sequencing combined with chromatin immunoprecipitation and a large variety of DNA/RNA-enrichment methodologies is at the origin of data resources of major importance. Indeed these resources, available for multiple genomes, represent the most comprehensive catalogue of (i) cell, development and signal transduction-specified patterns of binding sites for transcription factors (‘cistromes’) and for transcription and chromatin modifying machineries and (ii) the patterns of specific local post-translational modifications of histones and DNA (‘epigenome’) or of regulatory chromatin binding factors. In addition, (iii) the resources specifying chromatin structure alterations are emerging. Importantly, these types of “omics” datasets populate increasingly public repositories and provide highly valuable resources for the exploration of general principles of cell function in a multi-dimensional genome–transcriptome–epigenome–chromatin structure context. However, data mining is critically dependent on the data quality, an issue that, surprisingly, is still largely ignored by scientists and well-financed consortia, data repositories and scientific journals. So what determines the quality of ChIP-seq experiments and the datasets generated therefrom and what refrains scientists from associating quality criteria to their data? In this ‘opinion’ we trace the various parameters that influence the quality of this type of datasets, as well as the computational efforts that were made until now to qualify them. Moreover, we describe a universal quality control (QC) certification approach that provides a quality rating for ChIP-seq and enrichment-related assays. The corresponding QC tool and a regularly updated database, from which at present the quality parameters of more than 8000 datasets can be retrieved, are freely accessible at www.ngs-qc.org.

Tumor mutational burden (TMB) is considered a potential biomarker for predicting the response and effect of immune checkpoint inhibitors (ICIs). However, there are still inconsistent standards of gene panels using next-generation sequencing and poor correlation between the TMB genes, immune cell infiltrating, and prognosis. We applied text-mining technology to construct specific TMB-associated gene panels cross various cancer types. As a case exploration, Pearson's correlation between TMB genes and immune cell infiltrating was further analyzed in colorectal cancer. We then performed LASSO Cox regression to construct a prognosis predictive model and calculated the risk score of each sample for receiver operating characteristic (ROC) analysis. The results showed that the assessment of TMB gene panels performed well with fewer than 500 genes, highly mutated genes, and the inclusion of synonymous mutations and immune regulatory and drug-target genes. Moreover, the analysis of TMB differentially expressed genes (DEGs) suggested that JAKMIP1 was strongly correlated with the gene expression level of CD8+ T cell markers in colorectal cancer. Additionally, the prognosis predictive model based on 19 TMB DEGs reached AUCs of 0.836, 0.818, and 0.787 in 1-, 3-, and 5-year OS models, respectively (C-index: 0.810). In summary, the gene panel performed well and TMB DEGs showed great potential value in immune cell infiltration and in predicting survival. [ABSTRACT FROM AUTHOR]