Search

Biases in data used to train machine learning (ML) models can inflate their prediction performance and confound our understanding of how and what they learn. Although biases are common in biological data, systematic auditing of ML models to identify and eliminate these biases is not a common practice when applying ML in the life sciences. Here we devise a systematic, principled, and general approach to audit ML models in the life sciences. We use this auditing framework to examine biases in three ML applications of therapeutic interest and identify unrecognized biases that hinder the ML process and result in substantially reduced model performance on new datasets. Ultimately, we show that ML models tend to learn primarily from data biases when there is insufficient signal in the data to learn from. We provide detailed protocols, guidelines, and examples of code to enable tailoring of the auditing framework to other biomedical applications., Fatma-Elzahraa Eid et al. illustrate a principled approach for identifying biases that can inflate the performance of biological machine learning models. When applied to three biomedical prediction problems, they identify previously unrecognized biases and ultimately show that models are likely to learn primarily from data biases when there is insufficient learnable signal in the data.

COVID-19 CG (covidcg.org) is an open resource for tracking SARS-CoV-2 single-nucleotide variations (SNVs), lineages, and clades using the virus genomes on the GISAID database while filtering by location, date, gene, and mutation of interest. COVID-19 CG provides significant time, labor, and cost-saving utility to projects on SARS-CoV-2 transmission, evolution, diagnostics, therapeutics, vaccines, and intervention tracking. Here, we describe case studies in which users can interrogate (1) SNVs in the SARS-CoV-2 spike receptor binding domain (RBD) across different geographical regions to inform the design and testing of therapeutics, (2) SNVs that may impact the sensitivity of commonly used diagnostic primers, and (3) the emergence of a dominant lineage harboring an S477N RBD mutation in Australia in 2020. To accelerate COVID-19 efforts, COVID-19 CG will be upgraded with new features for users to rapidly pinpoint mutations as the virus evolves throughout the pandemic and in response to therapeutic and public health interventions.The discovery of faster spreading variants of the virus that causes coronavirus disease 2019 (COVID-19) has raised alarm. These new variants are the result of changes (called mutations) in the virus’ genetic code. Random mutations can occur each time a virus multiplies. Although most mutations do not introduce any meaningful changes, some can alter the characteristics of the virus, for instance, helping the virus to spread more easily, reinfecting people who have had COVID-19 before, or reducing the sensitivity to treatments or vaccines. Scientists need to know about mutations in the virus that make treatments or vaccines less effective as soon as possible, so they can adjust their pandemic response. As a result, tracking these genetic changes is essential. But individual scientists or public health agencies may not have the staff, time or computer resources to extract usable information from the growing amount of genetic data available. A free online tool created by Chen et al. may help scientists and public health officials to track changes to the virus more easily. The COVID-19 CoV Genetics tool (COVID-19 CG) can quickly provide information on which virus mutations are present in an area during a specific period. It does this by processing data on mutations found in viral genetic material collected worldwide from hundreds of thousands of people with COVID-19, which are hosted in an existing online database. The COVID-19 CG tool presents customizable, interactive visualizations of the data. Thousands of scientists, public health agencies, and COVID-19 vaccine and treatment developers in over 100 countries are already using the COVID-19 CG tool to find the most common mutations in their area and use it for research. They can use this information to develop more effective vaccines or treatments. Chen et al. plan to update and improve the tool as more information becomes available to help advance global efforts to end the COVID-19 pandemic.

Oxidative stress alters cell viability, from microorganism irradiation sensitivity to human aging and neurodegeneration. Deleterious effects of protein carbonylation by reactive oxygen species (ROS) make understanding molecular properties determining ROS-susceptibility essential. The radiation-resistant bacterium Deinococcus radiodurans accumulates less carbonylation than sensitive organisms, making it a key model for deciphering properties governing oxidative stress resistance. We integrated shotgun redox proteomics, structural systems biology, and machine learning to resolve properties determining protein damage by γ-irradiation in Escherichia coli and D. radiodurans at multiple scales. Local accessibility, charge, and lysine enrichment accurately predict ROS-susceptibility. Lysine, methionine, and cysteine usage also contribute to ROS-resistance of the D. radiodurans proteome. Our model predicts proteome maintenance machinery and proteins protecting against ROS are more resistant in D. radiodurans. Our findings substantiate that protein-intrinsic protection impacts oxidative stress resistance, identifying causal molecular properties.One Sentence SummaryProteins differ in intrinsic susceptibility to oxidation, a mode of evolutionary adaptation for stress tolerance in bacteria.

SUMMARYEctopic R-loop accumulation causes DNA replication stress and genome instability. To avoid these outcomes, cells possess a range of anti-R-loop mechanisms, including RNaseH that degrades the RNA moiety in R-loops. To comprehensively identify anti-R-loop mechanisms, we performed a genome-wide trigenic interaction screen in yeast lacking RNH1 and RNH201. We identified >100 genes critical for fitness in the absence of RNaseH, which were enriched for DNA replication fork maintenance factors such as RAD50. We show in yeast and human cells that R-loops accumulate during RAD50 depletion. In human cancer cell models, we find that RAD50 and its partners in the MRE11-RAD50-NBS1 complex regulate R-loop-associated DNA damage and replication stress. We show that a non-nucleolytic function of MRE11 is important for R-loop suppression via activation of PCNA-ubiquitination by RAD18 and recruiting anti-R-loop helicases in the Fanconi Anemia pathway. This work establishes a novel role for MRE11-RAD50-NBS1 in directing tolerance mechanisms of transcription-replication conflicts.

Genome recoding will provide a deeper understanding of genetics and transform biotechnology. We bypass the reliance of previous genome recoding methods on site-specific enzymes and demonstrate a rapid recombineering based strategy for writing genomes by Stepwise Integration of Rolling Circle Amplified Segments (SIRCAS). We installed the largest number of codon substitutions in a single organism yet published, creating a strain of Salmonella typhimurium with 1557 leucine codon changes across 200 kb of the genome.

DNA:RNA hybrid formation is emerging as a significant cause of genome instability in biological systems ranging from bacteria to mammals. Here we describe the genome-wide distribution of DNA:RNA hybrid prone loci in Saccharomyces cerevisiae by DNA:RNA immunoprecipitation (DRIP) followed by hybridization on tiling microarray. These profiles show that DNA:RNA hybrids preferentially accumulated at rDNA, Ty1 and Ty2 transposons, telomeric repeat regions and a subset of open reading frames (ORFs). The latter are generally highly transcribed and have high GC content. Interestingly, significant DNA:RNA hybrid enrichment was also detected at genes associated with antisense transcripts. The expression of antisense-associated genes was also significantly altered upon overexpression of RNase H, which degrades the RNA in hybrids. Finally, we uncover mutant-specific differences in the DRIP profiles of a Sen1 helicase mutant, RNase H deletion mutant and Hpr1 THO complex mutant compared to wild type, suggesting different roles for these proteins in DNA:RNA hybrid biology. Our profiles of DNA:RNA hybrid prone loci provide a resource for understanding the properties of hybrid-forming regions in vivo, extend our knowledge of hybrid-mitigating enzymes, and contribute to models of antisense-mediated gene regulation. A summary of this paper was presented at the 26th International Conference on Yeast Genetics and Molecular Biology, August 2013., Author Summary RNA processing factors are mutated in human cancers, inherited developmental disorders and neurodegenerative syndromes. Defects in RNA processing have been associated with increased levels of mutations and DNA damage in part via the formation of DNA:RNA hybrids. Although it is likely that specific regions of the genome are more prone to DNA:RNA hybrid formation, a map of hybrid-prone regions is not available. In this study, we describe the genome-wide distribution of DNA:RNA hybrids in both normal and mutant Saccharomyces cerevisiae cells. The resulting profiles contribute to both our understanding of the general properties of hybrid-forming loci and to our knowledge of hybrid-mitigating enzymes. Interestingly, significant DNA:RNA hybrid enrichment was detected at genes associated with antisense transcription. We show that overexpression of RNase H, which degrades the RNA in hybrids, significantly affects the expression of genes associated with antisense transcripts. These findings support a role for DNA:RNA hybrids in regulation of gene expression by antisense transcripts.