Celvion
Celvion provides EMphora, a combination design engine that predicts which multi-target regimens will shut down escape routes in solid tumors, where 90% of cancer deaths occur. The platform is trained exclusively on combination regimen data and validated against held-out trials, time-split trials, and 1,000+ real-world NSCLC patients, achieving 0.89 AUC on clinical trial cohorts. EMphora is currently applied in partnership with Rakovina Therapeutics on its ATR inhibitor program.
- Drug Discovery & Therapeutics
- Artificial Intelligence
- Healthcare Technology
- Biotechnology
- Data & Analytics
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Combination therapy is standard in oncology, yet choosing the right targets remains largely guesswork. The number of viable multi-target combinations far exceeds what any preclinical screen or clinical trial program can test one at a time, and preclinical synergy metrics often fail to translate to patient benefit due to missing clinical context such as resistance history, toxicity, and tumor evolution. Solid tumors frequently escape single-agent pressure through redundant pathways, lineage plasticity, and immune evasion, leaving pharma without a systematic method to identify promising combinations before committing to expensive trials.
Solution
Celvion provides EMphora, a design engine that predicts which target combinations shut down escape routes in solid tumors, where 90% of cancer deaths occur. The platform ingests clinical trial and real-world outcomes integrated with mechanistic and biological evidence, then scores combinations of two or more targets by predicted clinical benefit and surfaces the patient subgroups most likely to respond. EMphora is trained specifically on combination regimens rather than single agents, which the company demonstrates is essential for accurate predictions. The system is benchmarked against real clinical outcomes using held-out trials, time-split trials, and real-world patient data, and its outputs are designed to serve as confirmatory studies before a program advances. The platform is currently applied in partnership with Rakovina Therapeutics on its ATR inhibitor program to identify combination targets, predictive biomarkers, and patient populations most likely to benefit.
Target Audience
Primary customers are oncology drug developers and biopharmaceutical companies seeking to prioritize combination regimens before committing to expensive clinical trials, as well as clinical development teams needing data-driven rationale for target and biomarker selection.
Features
- Outcome-trained predictive models scoring combinations of 2+ targets by predicted clinical benefit, trained exclusively on combination regimen data rather than single-agent data
- Validation against held-out trials, time-split trials (trained on past, tested on later), and 1,000+ real-world NSCLC patients
- Performance metrics include 0.89 AUC on clinical trial cohorts, 0.88 AUC on time-split trials, 0.80 AUC on real-world NSCLC patients, and 0.87 Precision@5
- 2.06x lift in true responders for top 5 picks versus random selection
- Patient subgroup identification to surface populations most likely to respond to specific combinations
- Integration of clinical outcomes, real-world data, and mechanistic/biological evidence into a unified ranking framework