With cancer responsible for one in six deaths around the world, precision medicines based on the root cause of the disease, genetic mutations, are vital. Developing these targeted therapies depends on identifying the right targets – specific genes that tumour cells rely on for survival or proliferation so that, when silenced or knocked out, cancer cells are killed without harming healthy tissue.
The Cancer Dependency Map, or DepMap, and the Open Targets Platform are both examples of large-scale collaborative research initiatives aimed at accelerating this approach. Through systematic use of CRISPR-Cas9 and RNA interference, the team behind DepMap disabled every gene inside over 900 cancer cell lines and tracked which knockouts reduced viability or growth across the cells. This shows which genes are essential for tumour survival – information that was made public in 2018 to accelerate research.
Today, these databases are widely used in target identification and biomarker discovery. However, while strong in their own regard, they lack interpretability, providing no insight into mechanism of action, and cannot support researchers with ranking ideas that promise the greatest clinical upside.
There are also gaps in the data. Neither platform covers complex perturbations or vulnerabilities that only appear in specific molecular contexts – and the more we learn about cancer, the more we understand just how complex, heterogenous, and dynamic tumour biology can be. One example is the presence of synthetic lethal pairs – instances where the loss of one genetic dependency makes cancer cells reliant on a second DNA damage response (DDR) pathway for survival. By identifying and understanding these complexities, researchers can uncover hidden vulnerabilities that lead to exciting new targets or predictive biomarkers.
This is part of the mission behind Virtual Assays, machine learning applications that model cancer cells across hundreds of cell lines to predict genetic dependencies and post-perturbation RNA sequencing responses in silico. Synthetic lethality training is at the heart of this approach.
Synthetic lethal approaches in action
In the most well-studied example of synthetic lethality, certain cancers with pre-existing BRCA1/2 mutations are unable to repair DNA damage and turn to the PARP pathway – their ‘backup’ DDR mechanism – for survival. Through understanding of this dependency, researchers have developed synthetic lethal agents (in this case, PARP inhibitors) that block it, meaning cancer cells suffer while normal cells, with intact BRCA genes, are unaffected.
While this approach has transformed treatment for BRCA-mutant cancers, about 40% of patients still do not respond to PARP inhibitors, suggesting better ways to predict response are needed. To address this, patient-representative cell line models were built and screened for drug effects in Turbine’s Virtual Lab. The team simulated roughly 30,000 combinations of perturbations and PARP inhibitor concentrations across 25 sensitive and 37 resistant cell lines, spanning seven cancer types. By focusing on changes that altered response to PARP inhibitors, the experiments highlighted 13 potential biomarker candidates, and eight of these were validated in vitro. Two biomarkers are now included in clinical panels, including a novel biomarker unknown at the time, which is now used to guide patient selection in multiple PARP inhibitor clinical trials.
Finding novel alternatives to BRCA1/2
Resistance is a key limitation for oncology treatments. For PARP inhibitors, the challenge lies in the patient’s restoration of homologous recombination repair, which occurs in roughly 60% of cases. To explore novel solutions, Turbine used its platform to train a DDR-specific model, screening hundreds of simulated cancer cell lines to search for those that could be synthetically lethal with PARP-related mechanisms.
During this screening exercise, Turbine’s in silico platform predicted at least two dependent cell lines beyond BRCA1/2. In particular, the model identified that PARP inhibitor-resistant medulloblastoma and head and neck carcinoma (HSC-2) cell lines were dependent on a gene called NEK1, a dependency that has since been validated in vivo in HSC-2. These results support Turbine’s ability to uncover new dependencies and identify alternative targets for antitumour interventions, since this insight was not visible using DepMap alone.
The Turbine difference
Turbine’s virtual cell models harmonise more data modalities than any comparable platform, meaning they’re trained using high-throughput screening data like DepMap, as well as clinical and translational data, chemical and compound data, and specialised immune research data. Hosting fine-tuned virtual assays that guide or replace experiments, the Virtual Lab platform can run 50 million in silico experiments a day, investigating endless possibilities, ranking ideas, and building data packages so that only the strongest, derisked ideas are tested in the lab.
After ten years of development and collaborations with multiple large pharma organisations and research institutes, Turbine’s Virtual Assays and Virtual Lab was made accessible to clients in 2025. This year, the company is launching a new target discovery and de-risking suite that will help the industry identify novel targets and predictive biomarkers, developing first-in-class precision medicines that tie to clinical translatability.
It is an exciting time to be working in precision medicine. With powerful tools like DepMap as a starting point and Turbine’s Target Discovery & De-Risking module as the final arbiter, drug developers can now systematically link tumour biology to drug response at scale, identifying the right treatment for the right patients quickly and with confidence.
To learn more, read Turbine’s latest whitepaper on using Virtual Assays in the Virtual Lab by downloading below.
