The metabolome of an organism depends on environmental factors and intracellular regulation and provides information about the physiological conditions. Metabolomics helps to understand disease progression in clinical settings or estimate metabolite overproduction for metabolic engineering. The most popular analytical metabolomics platform is mass spectrometry (MS). However, MS metabolome data analysis is complicated, since metabolites interact nonlinearly, and the data structures themselves are complex. Machine learning methods have become immensely popular for statistical analysis due to the inherent nonlinear data representation and the ability to process large and heterogeneous data rapidly. In this review, we address recent developments in using machine learning for processing MS spectra and show how machine learning generates new biological insights. In particular, supervised machine learning has great potential in metabolomics research because of the ability to supply quantitative predictions. We review here commonly used tools, such as random forest, support vector machines, artificial neural networks, and genetic algorithms. During processing steps, the supervised machine learning methods help peak picking, normalization, and missing data imputation. For knowledge-driven analysis, machine learning contributes to biomarker detection, classification and regression, biochemical pathway identification, and carbon flux determination. Of important relevance is the combination of different omics data to identify the contributions of the various regulatory levels. Our overview of the recent publications also highlights that data quality determines analysis quality, but also adds to the challenge of choosing the right model for the data. Machine learning methods applied to MS-based metabolomics ease data analysis and can support clinical decisions, guide metabolic engineering, and stimulate fundamental biological discoveries.

An emergent area of cancer genomics has been the identification of driver genes. Driver genes confer a selective growth advantage to the cell and push it towards tumorigenesis. Functionally, driver genes can be divided into two categories, tumour suppressor genes (TSGs) and oncogenes (OGs), which have distinct mutation type profiles. While several driver genes have been discovered, many remain undiscovered, especially those that are mutated at a low frequency across samples. The current methods are not sufficient to predict all driver genes because the underlying characteristics of these genes are not yet well understood. Thus, to predict novel genes, we need to define new features and models that are not biased and identify genes that might otherwise be overshadowed by mutation profiles of recurrent driver genes. In this study, we define new features and build a model to identify novel driver genes. We overcome overfitting and show that certain mutation types such as nonsense mutations are more important for classification. Some known cancer driver genes, which are predicted by the model as TSGs with high probability are ARID1A, TP53, and RB1. In addition to these known genes, potential driver genes predicted are CD36, ZNF750 and ARHGAP35 as TSGs and TAB3 as an oncogene. Overall, our approach surmounts the issue of low recall and bias towards genes with high mutation rates and predicts potential novel driver genes for further experimental screening.

Despite the tremendous increase in omics data generated by modern sequencing technologies, their analysis can be tricky and often requires substantial expertise in bioinformatics. To address this concern, we have developed a user-friendly pipeline to analyze (cancer) genomic data that takes in raw sequencing data (FASTQ format) as input and outputs insightful statistics on the nature of the data. Our iCOMIC toolkit pipeline can analyze whole-genome and transcriptome data and is embedded in the popular Snakemake workflow management system. iCOMIC is characterized by a user-friendly GUI that offers several advantages, including executing analyses with minimal steps, eliminating the need for complex command-line arguments. The toolkit features many independent core workflows for both whole genomic and transcriptomic data analysis. Even though all the necessary, well-established tools are integrated into the pipeline to enable ‘out-of-the-box’ analysis, we provide the user with the means to replace modules or alter the pipeline as needed. Notably, we have integrated algorithms developed in-house for predicting driver and passenger mutations based on mutational context and tumor suppressor genes and oncogenes from somatic mutation data. We benchmarked our tool against Genome In A Bottle (GIAB) benchmark dataset (NA12878) and got the highest F1 score of 0.971 and 0.988 for indels and SNPs, respectively, using the BWA MEM - GATK HC DNA-Seq pipeline. Similarly, we achieved a correlation coefficient of r=0.85 using the HISAT2-StringTie-ballgown and STAR-StringTie-ballgown RNA-Seq pipelines on the human monocyte dataset (SRP082682). Overall, our tool enables easy analyses of omics datasets, with minimal steps, significantly ameliorating complex data analysis pipelines. Availability: https://github.com/RamanLab/iCOMIC

The progression of tumorigenesis starts with a few mutational and structural driver events in the cell. Various cohort-based computational tools exist to identify driver genes but require a large number of samples to produce reliable results. Many studies use different methods to identify driver mutations/genes from mutations that have no impact on tumour progression; however, a small fraction of patients show no mutational events in any known driver genes. Current unsupervised methods map somatic and expression data onto a network to identify the perturbation in the network. Our method is the first machine learning model to classify genes as tumour suppressor gene (TSG), oncogene (OG) or neutral, thus assigning the functional impact of the gene in the patient. In this study, we develop a multi-omic approach, PIVOT (Personalised Identification of driVer OGs and TSGs), to train on experimentally or computationally validated mutational and structural driver events. Given the lack of any gold standards for the identification of personalised driver genes, we label the data using four strategies and, based on classification metrics, show gene-based labelling strategies perform best. We build different models using SNV, RNA, and multi-omic features to be used based on the data available. Our models trained on multi-omic data improved predictions compared to mutation and expression data, achieving an accuracy ≥ 0.99 for BRCA, LUAD and COAD datasets. We show network and expression-based features contribute the most to PIVOT. Our predictions on BRCA, COAD and LUAD cancer types reveal commonly altered genes such as TP53, and PIK3CA, which are predicted drivers for multiple cancer types. Along with known driver genes, our models also identify new driver genes such as PRKCA, SOX9 and PSMD4. Our multi-omic model labels both CNV and mutations with a more considerable contribution by CNV alterations. While predicting labels for genes mutated in multiple samples, we also label rare driver events occurring in as few as one sample. We also identify genes with dual roles within the same cancer type. Overall, PIVOT labels personalised driver genes as TSGs and OGs and also identifies rare driver genes. PIVOT is available at https://github.com/RamanLab/PIVOT .