Biology subjects

Rossen, J.

Publications and source records attributed to Rossen, J..

4 recordsLinked to original sources

bioRxiv2026

Agentic systems are adept at solving well-scoped, verifiable problems in computational biology

We introduce CompBioBench, a benchmark of 100 diverse tasks for evaluating agentic systems in computational biology. Unlike mathematics and programming, which more readily admit systematic verification, biological data are inherently noisy and open to interpretation. To enable objective evaluation without reducing tasks to prescriptive checklists, we propose a new benchmark construction strategy based on synthetic/augmented data and metadata scrambling/scrubbing of real datasets to create challenging problems that have a single ground-truth answer and require multi-step reasoning, tool use, bespoke code, and interaction with real-world external resources. The benchmark spans genomics, transcriptomics, epigenomics, single-cell analysis, human genetics, and machine learning workflows. Questions are curated by domain experts to cover a broad range of skills with varying difficulty. We evaluate leading general-purpose agentic systems starting from a bare-minimum environment, requiring them to fetch data and tools as needed to solve each problem. We find strong end-to-end performance, with Codex CLI (GPT 5.4) reaching 83% accuracy, Gemini CLI (3.1 Pro) reaching 82%, Claude Code (Opus 4.6) reaching 81%, and Claude Code (Opus 4.7) reaching 78%. On the hardest questions, Claude Code (Opus 4.6) reaches 69%, Codex CLI (GPT 5.4) reaches 59%, and Gemini CLI (3.1 Pro) reaches 49%. CompBioBench provides a practical testbed for measuring the progress of agentic systems in computational biology and for guiding future benchmark design. Data and a public leaderboard are available at https://huggingface.co/collections/Genentech/compbiobench-v1.

bioinformatics↗

bioRxiv2019

Non-oncology drugs are a source of previously unappreciated anti-cancer activity

Anti-cancer uses of non-oncology drugs have been found on occasion, but such discoveries have been serendipitous and rare. We sought to create a public resource containing the growth inhibitory activity of 4,518 drugs tested across 578 human cancer cell lines. To accomplish this, we used PRISM, which involves drug treatment of molecularly barcoded cell lines in pools. Relative barcode abundance following treatment thus reflects cell line viability. We found that an unexpectedly large number of non-oncology drugs selectively inhibited subsets of cancer cell lines. Moreover, the killing activity of the majority of these drugs was predictable based on the molecular features of the cell lines. Follow-up of several of these compounds revealed novel mechanisms. For example, compounds that kill by inducing PDE3A-SLFN12 complex formation; vanadium-containing compounds whose killing is dependent on the sulfate transporter SLC26A2; the alcohol dependence drug disulfiram, which kills cells with low expression of metallothioneins; and the anti-inflammatory drug tepoxalin, whose killing is dependent on high expression of the multi-drug resistance gene ABCB1. These results illustrate the potential of the PRISM drug repurposing resource as a starting point for new oncology therapeutic development. The resource is available at https://depmap.org.

cancer biology↗

bioRxiv2019

Extracting Biological Insights from the Project Achilles Genome-Scale CRISPR Screens in Cancer Cell Lines

One of the main goals of the Cancer Dependency Map project is to systematically identify cancer vulnerabilities across cancer types to accelerate therapeutic discovery. Project Achilles serves this goal through the in vitro study of genetic dependencies in cancer cell lines using CRISPR/Cas9 (and, previously, RNAi) loss-of-function screens. The project is committed to the public release of its experimental results quarterly on the DepMap Portal (https://depmap.org), on a pre-publication basis. As the experiment has evolved, data processing procedures have changed. Here we present the current and projected Achilles processing pipeline, including recent improvements and the analyses that led us to adopt them, spanning data releases from early 2018 to the first quarter of 2020. Notable changes include quality control metrics, calculation of probabilities of dependency, and correction for screen quality and other biases. Developing and improving methods for extracting biologically-meaningful scores from Achilles experiments is an ongoing process, and we will continue to evaluate and revise data processing procedures to produce the best results.

cancer biology↗

bioRxiv2019

Surveillance-embedded genomic outbreak resolution of methicillin-susceptible Staphylococcus aureus in a neonatal intensive care unit

BackgroundWe observed an increase in methicillin-susceptible Staphylococcus aureus (MSSA) infections among neonates at a Dutch third level neonatal intensive care unit. Weekly surveillance data of MSSA carriage among neonates and cross-sectional screenings of health care workers (HCWs) were available for outbreak tracing. While traditional typing of MSSA isolates by staphylococcal protein A gene (spa typing) and Multiple-Locus Variable number tandem repeat Analysis (MLVA) suggested that nosocomial transmission had contributed to the infections, here they lacked the resolution to draw solid conclusions.\n\nMethodsMSSA isolates from neonatal infections, carriage surveillance, and HCWs were subjected to whole-genome sequencing and compared by a series of automated tools including de novo assembly, identification and localization of high-quality single nucleotide polymorphisms, and in-depth analysis of subsets of isolates. Outbreaks were defined as isolates that were more closely related than was to be expected from the genetic diversity in background surveillance.\n\nResultsGenomic analysis identified isolates that had been unjustly assigned to clusters based on MLVA typing, while spa typing was concordant but of insufficient resolution. Detailing particular subsets of isolates further improved resolution and although it provided evidence that HCWs were involved in multiple outbreaks, it alleviated heavy concerns about one particular HCW. Genomic clustering of isolates based on deviations from background surveillance matched epidemiological patient linkage. Compared to MLVA typing, the genomic analysis demonstrated more, shorter, and re-assorted nosocomial transmission chains during this outbreak.\n\nConclusionsIn this study the improved resolution and accuracy of genomic outbreak analyses compared to spa typing and MLVA substantially altered the view on outbreaks, along with apposite outbreak measures. Inclusion of the circulating background population has the potential to overcome current issues in genomic outbreak inference.

microbiology↗