PITCH DECK ONE-PAGER SEMIA / Quest for Health backs health ventures from Strasbourg across the full instrument stack, from the Bourse Grand Est to the EIC pathway, and the new Biotech-Pharma track is where this venture belongs. Drug resistance kills cancer and antiviral therapies before patients finish a course, and predicting which protein mutations break which drug currently depends on having a crystal structure. Structure-based tools cover roughly 17.6 percent of the mutations clinicians actually encounter, leaving the rest unscored. The venture predicts drug resistance mutations from protein sequence alone, with no crystal structure required. The model pairs ESM-2 protein language model delta-embeddings with ECFP4 drug fingerprints and a Random Forest classifier. On the Platinum benchmark of 357 mutations under protein-grouped cross-validation it reaches AUROC 0.634, against published mCSM-lig AUROC 0.70, and it does so at 100 percent mutation coverage versus 17.6 percent for structure-limited tools. Founder: Eniola Olutogun, pharmacist turned ML engineer, sole author of the proof-of-concept. Stage: pre-seed, proof-of-concept validated, not yet incorporated, planning incorporation in Île-de-France. The ask of SEMIA: incubation on the Biotech-Pharma track, the Bourse Grand Est at roughly 30K euros to fund the SKEMPI fine-tune, and introductions toward WiClub Santé and the i-Lab to EIC pathway. Named industry and academic targets within reach of the Grand Est and Paris-Saclay corridor: Servier in Suresnes, Sanofi in Gentilly, and I2BC and Institut Pasteur at Paris-Saclay. PITCH DECK SLIDES SLIDE 1 TITLE Predicting drug resistance mutations from protein sequence alone. Eniola Olutogun, pharmacist-turned-ML engineer. Pre-seed, proof-of-concept validated. Applying to the SEMIA / Quest for Health Biotech-Pharma track, Strasbourg. SLIDE 2 PROBLEM A single point mutation in a target protein can void an oncology or antiviral drug mid-treatment. Predicting these resistance mutations today needs a solved crystal structure of the protein-drug complex. Structure-based tools such as mCSM-lig score only about 17.6 percent of clinically relevant mutations, because the other structures do not exist. SLIDE 3 SOLUTION A sequence-only resistance predictor. Inputs are the wild-type and mutant protein sequence plus the drug. No structure is required, so coverage reaches 100 percent of mutations. The output is a per-mutation resistance probability usable in lead optimisation and in clinical resistance surveillance. SLIDE 4 TECHNOLOGY ESM-2 protein language model delta-embeddings capture the sequence change. ECFP4 fingerprints encode the drug. A Random Forest classifier maps the pair to a resistance call. The pipeline runs on commodity GPU and needs no docking or molecular dynamics. SLIDE 5 VALIDATION Platinum benchmark, 357 mutations, protein-grouped cross-validation to prevent leakage across homologous proteins. AUROC 0.634. Published structure-based SOTA mCSM-lig reaches AUROC 0.70 but only on the 17.6 percent of mutations it can score. The trade is a modest accuracy gap for nearly six times the coverage. SLIDE 6 ROADMAP Fine-tune ESM-2 on the SKEMPI set of roughly 3,000 mutations to lift AUROC to 0.70 or above. Run a pilot with Servier in Suresnes. Convert the pilot to annual recurring revenue. The Bourse Grand Est funds the fine-tune; WiClub Santé and i-Lab fund the pilot and incorporation; the EIC pathway funds scale-up. SLIDE 7 MARKET AND PARTNERS Buyers are pharma lead-optimisation and resistance-surveillance teams in oncology, antivirals, and antimicrobial resistance. Named targets: Servier (Suresnes), Sanofi (Gentilly), and academic validation partners I2BC and Institut Pasteur at Paris-Saclay. SLIDE 8 TEAM AND ASK Sole founder Eniola Olutogun, pharmacist with ML engineering depth, author of the full proof-of-concept. Ask of SEMIA: a place on the Biotech-Pharma track, the Bourse Grand Est at roughly 30K euros, and the staged pathway through WiClub Santé and i-Lab toward an EIC Accelerator application up to 2.5M euros. FOUNDING TEAM CV Eniola Olutogun, sole founder Profile: pharmacist trained in drug action who moved into machine learning engineering, now building and validating computational models end to end. Sole author of the venture's proof-of-concept, from dataset assembly through model architecture, cross-validation design, and benchmarking. Domain bridge: pharmacy training grounds the work in how resistance arises and matters clinically; ML engineering delivers the ESM-2 plus ECFP4 plus Random Forest pipeline that produces the AUROC 0.634 Platinum result. Technical work delivered: built the sequence-only resistance predictor; designed protein-grouped cross-validation to avoid homology leakage; benchmarked against mCSM-lig; quantified the coverage advantage at 100 percent versus 17.6 percent. Current execution: leading conversations with SEMIA / Quest for Health, meeting Romain Neidl on 29 June 2026. Parallel pipeline includes WILCO One BioTech in October 2026 and submitted applications to IncubAlliance and AI House. Location plan: incorporating in Île-de-France, positioned for the Paris-Saclay and Grand Est life-science corridor. [INSERT: degrees, institutions, and graduation years; prior employers and titles; ORCID and any preprint or publication links; location and work authorisation status.] TECHNICAL ONE-PAGER Problem Resistance mutations in drug-target proteins erode therapy in oncology, antivirals, and antimicrobial resistance. Established predictors such as mCSM-lig require a solved structure of the protein-ligand complex and therefore score only about 17.6 percent of mutations seen in practice. Approach The method predicts resistance from sequence alone. For each mutation it computes ESM-2 protein language model embeddings for wild-type and mutant sequences and takes their delta, encodes the drug with ECFP4 fingerprints, and classifies the concatenated representation with a Random Forest. No crystal structure, docking, or molecular dynamics is involved, which is what lifts coverage to 100 percent of mutations. Results On the Platinum benchmark, 357 mutations under protein-grouped cross-validation, the model reaches AUROC 0.634. Protein-grouped splits prevent the same protein family appearing in both train and test, so the figure reflects generalisation to unseen targets rather than memorisation. Published structure-based SOTA, mCSM-lig, reports AUROC 0.70 on the narrow structure-available subset. The result is competitive accuracy with roughly a six-fold gain in applicability. Next milestone Fine-tune ESM-2 on the SKEMPI set of approximately 3,000 binding-affinity mutations to push AUROC to 0.70 or above, matching structure-based accuracy at full coverage. This is the deliverable the Bourse Grand Est funds. Validation and commercial path Academic validation with I2BC and Institut Pasteur at Paris-Saclay. Commercial pilot with Servier in Suresnes, with Sanofi in Gentilly as a second target, converting to annual recurring revenue. The SEMIA instrument stack then carries the venture from Bourse Grand Est through WiClub Santé and i-Lab to an EIC Accelerator application. [INSERT: software stack and licences; data provenance and licensing for Platinum and SKEMPI; compute footprint; any IP or filing status.] CHECKLIST - [ ] Pitch deck one-pager finalised and exported to PDF - [ ] Pitch deck slides finalised and exported to PDF - [ ] Founding team CV completed with degrees, employers, ORCID, and links - [ ] Technical one-pager completed with stack, data licensing, and IP status - [ ] Confirm meeting with Romain Neidl on 29 June 2026 and prepare talking points on the Biotech-Pharma track - [ ] Confirm Bourse Grand Est eligibility criteria and required supporting documents with SEMIA - [ ] Prepare WiClub Santé and i-Lab pathway questions for the SEMIA meeting - [ ] Verify data licences for Platinum and SKEMPI before any commercial pilot - [ ] Decide and document incorporation timing and entity location in Île-de-France or Grand Est EDITOR NOTES - Eligibility: the profile states no gate and a meeting already in progress, but the Bourse Grand Est is a Grand Est regional instrument while incorporation is planned in Île-de-France; confirm whether a Grand Est presence or registration is required to draw that 30K, since it affects the whole staged pathway. - Verify the numbers before submission: AUROC 0.634 on Platinum, 357 mutations, 17.6 percent versus 100 percent coverage, mCSM-lig 0.70, and SKEMPI at roughly 3,000 mutations should each be source-checked and dated. - The CV is a skeleton; the applicant must insert real degrees, institutions, graduation years, prior employers, ORCID, and any preprint links, since none are in the profile. - Confirm the relationships with Servier, Sanofi, I2BC, and Institut Pasteur are genuine targets versus active conversations, and phrase them honestly in the live meeting; do not let the deck imply commitments that do not exist. - The "pharmacist-turned-ML engineer" framing needs at least one verifiable credential or shipped project on each side to be credible to a health incubator; add specifics before the 29 June meeting.