Draft — MSK Therapeutics Accelerator

MOTIVATION LETTER

The MSK Therapeutics Accelerator selects companies that can change how cancer is treated. My venture predicts drug resistance mutations from protein sequence alone, without requiring a crystal structure. On the Platinum benchmark of 553 mutations, the platform achieves an AUROC of 0.804 plus or minus 0.025 using protein-grouped cross-validation. This beats the published SOTA for mCSM-lig at roughly 0.70. More critically, the platform covers 100 percent of mutations in a given protein target, while structure-limited tools cover approximately 18 percent. For rare and undrugged cancers where no crystal structure exists, that coverage gap is the difference between a viable drug and a clinical failure.

MSK treats patients with cancers that have no standard-of-care options. Those patients often harbor resistance mutations that emerge during therapy. My platform can take a protein sequence from a patient biopsy and predict, within hours, which approved or pipeline drugs will fail and which combinations might succeed. The technology uses ESM-2 protein language model delta-embeddings fused with ECFP4 drug fingerprints, classified by a Random Forest. It is pre-seed, proof-of-concept validated, and not yet incorporated.

I am applying to this accelerator because MSK's clinical expertise and core facilities can validate the platform on MSK-relevant targets. I propose a co-development project: apply the prediction pipeline to a set of resistance mutations observed in MSK clinical trials, compare predictions against patient outcomes, and produce a validated assay that MSK can use to guide combination therapy design. This aligns with the accelerator's criteria for feasibility of co-development at MSK and alignment with MSK's expertise.

The venture is targeting EU incorporation in Ile-de-France, with named partners including Servier in Suresnes, Paris-Saclay I2BC, Institut Pasteur, and Sanofi in Gentilly. A US accelerator placement would open parallel validation pathways with MSK's network and create a transatlantic regulatory and clinical strategy. The platform is drug-agnostic and cancer-type-agnostic, but the most immediate impact is in oncology where resistance drives relapse.

I am a pharmacist turned machine learning engineer. I built the platform as sole author. I have a meeting in progress with SEMIA and Quest for Health, applications submitted to IncubAlliance and AI House, and a WILCO One BioTech slot reserved for October 2026. The MSK Therapeutics Accelerator is the right environment to pressure-test the platform against real clinical resistance data and turn a computational benchmark into a therapeutic tool.

SHORT ESSAY: PLATFORM INNOVATION AND CANCER IMPACT

The platform solves a specific bottleneck in cancer drug development: predicting which mutations will confer resistance before they appear in the clinic. Current methods require a co-crystal structure of the drug-protein complex. For 82 percent of clinically relevant mutations, that structure does not exist. My platform uses only the protein sequence and the drug's chemical fingerprint. On the SKEMPI 2.0 benchmark of binding affinity changes, the platform achieves an AUROC of 0.634. The roadmap targets fine-tuning ESM-2 on 3,000 SKEMPI mutations to reach an AUROC of 0.70 or higher, which would match or exceed structure-based methods on their own benchmarks.

The cancer impact is direct. A drug that passes Phase I but fails in Phase II because of unanticipated resistance costs an average of 1.2 billion euros in sunk development. My platform can be deployed at the preclinical stage to screen a drug candidate against the full mutational landscape of its target, flagging resistance liabilities before animal studies begin. At the clinical stage, it can interpret patient sequencing data to recommend second-line therapies. MSK's patient population includes heavily pretreated individuals where resistance mechanisms are complex and multi-drug. The platform's ability to cover 100 percent of mutations without structural input makes it uniquely suited to those cases.

SHORT ESSAY: CO-DEVELOPMENT PLAN WITH MSK

The proposed co-development project has three phases. Phase one, months one to three: MSK identifies a set of 20 to 50 resistance mutations observed in its clinical trials for a specific target, for example EGFR in non-small cell lung cancer or KRAS G12C in colorectal cancer. My platform predicts resistance scores for all approved and pipeline drugs against those mutations. Phase two, months four to six: MSK's core facilities compare predictions against patient outcome data from MSK-IMPACT or similar sequencing cohorts. Discrepancies are analyzed to retrain the model. Phase three, months seven to twelve: the validated prediction pipeline is packaged as a web-based assay that MSK clinicians can query for any patient mutation. The deliverable is a published benchmark on MSK data and a tool integrated into MSK's clinical decision support workflow.

This plan is feasible because the platform is already built and validated on public benchmarks. The compute requirement is modest: a single GPU can run the full pipeline for a protein target in under two hours. MSK provides the clinical data and the domain expertise. I provide the model and the engineering. The outcome is a resistance-prediction assay that MSK can use internally and, if successful, license or co-develop for broader distribution.

CHECKLIST

- [ ] Motivation letter, 300-500 words, tailored to MSK Therapeutics Accelerator
- [ ] Short essay on platform innovation and cancer impact, 200-350 words
- [ ] Short essay on co-development plan with MSK, 200-350 words
- [ ] CV or resume of Eniola Olutogun
- [ ] Proof-of-concept validation data summary (AUROC 0.804 on Platinum, 0.634 on SKEMPI 2.0)
- [ ] List of named partners and support pipeline (Servier, Paris-Saclay, Institut Pasteur, Sanofi, SEMIA, WILCO, IncubAlliance, AI House)
- [ ] Technology stack description (ESM-2, ECFP4, Random Forest)
- [ ] Roadmap with milestones (fine-tune ESM-2 on 3K SKEMPI mutations, AUROC 0.70, Servier pilot, ARR)
- [ ] Confirmation of eligibility for US-based accelerator as non-US founder
- [ ] Application form on MSK Therapeutics Accelerator website

EDITOR NOTES

- Eligibility risk: the accelerator is US-based and the venture is targeting EU incorporation. Confirm whether MSK accepts non-US founders and whether the accelerator requires a US entity. If yes, plan for Delaware C-corp or US subsidiary.
- Fact verification: the Platinum benchmark AUROC of 0.804 and mCSM-lig AUROC of 0.70 are drawn from the profile. Confirm these numbers are from a published preprint or internal validation report that can be shared with MSK reviewers.
- Gap: the profile does not specify which cancer types or drug targets the platform has been tested on beyond the generic benchmarks. The applicant should insert 2-3 specific examples, e.g., EGFR T790M, KRAS G12C, or BCR-ABL T315I, to ground the co-development plan in MSK's clinical reality.

MOTIVATION LETTER

The MSK Therapeutics Accelerator selects companies that can change how cancer is treated. My venture predicts drug resistance mutations from protein sequence alone, without requiring a crystal structure. On the Platinum benchmark of 553 mutations, the platform achieves an AUROC of 0.804 plus or minus 0.025 using protein-grouped cross-validation. This beats the published SOTA for mCSM-lig at roughly 0.70. More critically, the platform covers 100 percent of mutations in a given protein target, while structure-limited tools cover approximately 18 percent. For rare and undrugged cancers where no crystal structure exists, that coverage gap is the difference between a viable drug and a clinical failure.

MSK treats patients with cancers that have no standard-of-care options. Those patients often harbor resistance mutations that emerge during therapy. My platform can take a protein sequence from a patient biopsy and predict, within hours, which approved or pipeline drugs will fail and which combinations might succeed. The technology uses ESM-2 protein language model delta-embeddings fused with ECFP4 drug fingerprints, classified by a Random Forest. It is pre-seed, proof-of-concept validated, and not yet incorporated.

I am applying to this accelerator because MSK's clinical expertise and core facilities can validate the platform on MSK-relevant targets. I propose a co-development project: apply the prediction pipeline to a set of resistance mutations observed in MSK clinical trials, compare predictions against patient outcomes, and produce a validated assay that MSK can use to guide combination therapy design. This aligns with the accelerator's criteria for feasibility of co-development at MSK and alignment with MSK's expertise.

The venture is targeting EU incorporation in Ile-de-France, with named partners including Servier in Suresnes, Paris-Saclay I2BC, Institut Pasteur, and Sanofi in Gentilly. A US accelerator placement would open parallel validation pathways with MSK's network and create a transatlantic regulatory and clinical strategy. The platform is drug-agnostic and cancer-type-agnostic, but the most immediate impact is in oncology where resistance drives relapse.

I am a pharmacist turned machine learning engineer. I built the platform as sole author. I have a meeting in progress with SEMIA and Quest for Health, applications submitted to IncubAlliance and AI House, and a WILCO One BioTech slot reserved for October 2026. The MSK Therapeutics Accelerator is the right environment to pressure-test the platform against real clinical resistance data and turn a computational benchmark into a therapeutic tool.

SHORT ESSAY: PLATFORM INNOVATION AND CANCER IMPACT

The platform solves a specific bottleneck in cancer drug development: predicting which mutations will confer resistance before they appear in the clinic. Current methods require a co-crystal structure of the drug-protein complex. For 82 percent of clinically relevant mutations, that structure does not exist. My platform uses only the protein sequence and the drug's chemical fingerprint. On the SKEMPI 2.0 benchmark of binding affinity changes, the platform achieves an AUROC of 0.634. The roadmap targets fine-tuning ESM-2 on 3,000 SKEMPI mutations to reach an AUROC of 0.70 or higher, which would match or exceed structure-based methods on their own benchmarks.

The cancer impact is direct. A drug that passes Phase I but fails in Phase II because of unanticipated resistance costs an average of 1.2 billion euros in sunk development. My platform can be deployed at the preclinical stage to screen a drug candidate against the full mutational landscape of its target, flagging resistance liabilities before animal studies begin. At the clinical stage, it can interpret patient sequencing data to recommend second-line therapies. MSK's patient population includes heavily pretreated individuals where resistance mechanisms are complex and multi-drug. The platform's ability to cover 100 percent of mutations without structural input makes it uniquely suited to those cases.

SHORT ESSAY: CO-DEVELOPMENT PLAN WITH MSK

The proposed co-development project has three phases. Phase one, months one to three: MSK identifies a set of 20 to 50 resistance mutations observed in its clinical trials for a specific target, for example EGFR in non-small cell lung cancer or KRAS G12C in colorectal cancer. My platform predicts resistance scores for all approved and pipeline drugs against those mutations. Phase two, months four to six: MSK's core facilities compare predictions against patient outcome data from MSK-IMPACT or similar sequencing cohorts. Discrepancies are analyzed to retrain the model. Phase three, months seven to twelve: the validated prediction pipeline is packaged as a web-based assay that MSK clinicians can query for any patient mutation. The deliverable is a published benchmark on MSK data and a tool integrated into MSK's clinical decision support workflow.

This plan is feasible because the platform is already built and validated on public benchmarks. The compute requirement is modest: a single GPU can run the full pipeline for a protein target in under two hours. MSK provides the clinical data and the domain expertise. I provide the model and the engineering. The outcome is a resistance-prediction assay that MSK can use internally and, if successful, license or co-develop for broader distribution.

CHECKLIST

- [ ] Motivation letter, 300-500 words, tailored to MSK Therapeutics Accelerator
- [ ] Short essay on platform innovation and cancer impact, 200-350 words
- [ ] Short essay on co-development plan with MSK, 200-350 words
- [ ] CV or resume of Eniola Olutogun
- [ ] Proof-of-concept validation data summary (AUROC 0.804 on Platinum, 0.634 on SKEMPI 2.0)
- [ ] List of named partners and support pipeline (Servier, Paris-Saclay, Institut Pasteur, Sanofi, SEMIA, WILCO, IncubAlliance, AI House)
- [ ] Technology stack description (ESM-2, ECFP4, Random Forest)
- [ ] Roadmap with milestones (fine-tune ESM-2 on 3K SKEMPI mutations, AUROC 0.70, Servier pilot, ARR)
- [ ] Confirmation of eligibility for US-based accelerator as non-US founder
- [ ] Application form on MSK Therapeutics Accelerator website

EDITOR NOTES

- Eligibility risk: the accelerator is US-based and the venture is targeting EU incorporation. Confirm whether MSK accepts non-US founders and whether the accelerator requires a US entity. If yes, plan for Delaware C-corp or US subsidiary.
- Fact verification: the Platinum benchmark AUROC of 0.804 and mCSM-lig AUROC of 0.70 are drawn from the profile. Confirm these numbers are from a published preprint or internal validation report that can be shared with MSK reviewers.
- Gap: the profile does not specify which cancer types or drug targets the platform has been tested on beyond the generic benchmarks. The applicant should insert 2-3 specific examples, e.g., EGFR T790M, KRAS G12C, or BCR-ABL T315I, to ground the co-development plan in MSK's clinical reality.

AI Draft — MSK Therapeutics Accelerator