Background: Looking beyond gene pairs to cancer-cell context

Researchers developed DepMine, an analysis tool that combines cancer-cell conditions—including mutations, copy-number alterations, and gene expression—to identify candidate therapeutic targets. Using public cancer cell-line data, the team analyzed how specific genetic backgrounds relate to gene dependencies and presented a way to refine candidate biomarkers for target discovery.

Gene dependency describes how much a cell’s survival is affected when it loses the function of a particular gene. Large-scale loss-of-function screens using CRISPR or small interfering RNA (siRNA) are key ways to find these relationships, but analyzing the associated genomic information requires substantial data-science and coding expertise. Approaches that examine only pairs of genes can also struggle to explain why dependency differs among cancer cells carrying the same mutation.

Synthetic lethality is a relationship in which a cell can survive the loss of function of either of two genes individually but has difficulty surviving the loss of function of both. DepMine extends this analysis from individual gene pairs to groups of cancer cells defined by multiple conditions.

Key findings: Cancer profiles defined with set operations

Implemented in Python, DepMine selects cell lines based on conditions including mutations, copy-number alterations, expression levels, tissue of origin, and mutational signatures. Users can combine union, intersection, and exclusion operations to build cancer profiles, then compare gene dependencies between the selected group and a control group. The tool reduces the burden of writing complex analysis code, though users still need to define their questions using basic set-algebra expressions.

DepMine’s default data are based primarily on public DepMap 2023Q2 data. In a separate comprehensive pairwise analysis using DepMap 2022Q2, the researchers reported 34,256 candidate pairs among 16,825 genes after excluding genes with high mean dependency, among other exclusions. These are relationships detected using those data and statistical criteria; they are not all experimentally validated synthetic-lethal pairs.

As an example of composite profiling, the researchers analyzed microsatellite instability (MSI). A profile reflecting defects in DNA mismatch repair and proofreading during replication, among other conditions, included 142 cell lines. In the dependency comparison, WRN and RPL22L1 were statistically significant and had Cohen’s d, an effect-size measure, greater than 0.8. Rediscovering WRN, a known MSI-related target, supports the analysis’s validity, but this should be distinguished from validation of patient treatment responses.

Meaning and outlook: Balancing effect size and scope

DepMine searches for secondary factors that separate groups with high and low dependency within an initial profile. It then uses Pareto optimization to compare trade-offs between effect size and the number of included cell lines. In an example analyzing APC loss of function and CTNNB1 dependency, one selected profile incorporated conditions including tissue, age, and CDKN2A deletion and included 83 cell lines. Another option produced a larger effect size but reduced coverage to 62 cell lines. Researchers can choose whether to study a broader cancer-cell population or focus on a group with more pronounced dependency.

The study’s contribution is an analytical framework that supports the exploration of target and biomarker hypotheses in cell lines. Dependency measured by gene loss of function does not always match drug sensitivity, and increasingly refined profiles may narrow the analysis group too much. Using optimized profiles as patient-selection criteria will require separate experimental and clinical validation. The software and related materials are available free of charge to academic and nonprofit institutions.

The methods and results are available in the Bioinformatics paper.