MOTIVATION LETTER Drug resistance is the single largest cause of treatment failure in oncology and infectious disease. Predicting which mutations will confer resistance before a patient relapses or a pathogen escapes a drug is a computational problem that current tools solve for only eighteen percent of mutations because they require a protein crystal structure. My venture eliminates that bottleneck. Using ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints and a Random Forest classifier, the platform predicts resistance mutations from protein sequence alone. On the Platinum benchmark of 553 mutations, the model achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation, outperforming the published state-of-the-art mCSM-lig which scores approximately 0.70. Mutation coverage is one hundred percent versus eighteen percent for structure-dependent methods. I am a pharmacist and machine learning engineer, sole author of this venture, and a Nigerian citizen. The Commonwealth Startup Fellowship is the right programme because it targets entrepreneurs from low- and middle-income Commonwealth countries who have moved beyond idea stage into active business development. My venture meets that bar. The proof-of-concept is complete. The minimum viable product, the ESM-2 plus Random Forest classifier, is functional and benchmarked. I am in active pilot discussions with Servier in Suresnes and have established partnerships with Paris-Saclay, the I2BC, and Institut Pasteur. The next step is incorporation and securing pilot contracts that will generate the first annual recurring revenue. The fellowship bootcamp in Accra from 14 to 30 November 2025 offers a direct path to the networks I need: biopharma partnerships in Europe, compute credits for scaling the model on the SKEMPI 3K mutation dataset, and mentorship on incorporating in France or the European Union. I am available for the full schedule. I hold a graduate degree, am fluent in written and spoken English, have not received a Commonwealth Professional Fellowship in the last five years, and am a permanent resident of Nigeria. The venture also creates a specific opportunity for Nigeria. My plan is to build a distributed AI team in Lagos and Abuja, training Nigerian machine learning engineers on protein language model fine-tuning and drug discovery pipelines. This fellowship would accelerate that timeline by connecting me to European biopharma partners who can serve as pilot customers and to investors who understand the AI-biotech thesis. SHORT ESSAY: INNOVATION IN PRODUCT, SERVICE, OR MARKET STRATEGY The innovation is a structural decoupling of resistance prediction from crystallography. Every competing tool in this space, including mCSM-lig, DeepDDI, and DrugCell, requires a three-dimensional protein structure as input. That requirement excludes roughly eighty-two percent of clinically relevant mutations because the corresponding protein has no solved crystal structure. My venture uses ESM-2, a protein language model trained on 250 million sequences, to extract evolutionary information from sequence alone. Delta-embeddings capture the change in the protein's learned representation when a mutation is introduced, and the Random Forest classifier maps that change to a resistance probability for a given drug fingerprint. The market strategy targets a specific pain point. Pharmaceutical companies running preclinical drug discovery screen compounds against panels of resistance mutations. They currently outsource this to contract research organizations that perform time-consuming mutagenesis assays, or they accept the eighteen percent coverage of structure-based tools. My venture offers full coverage at a fraction of the cost and time. The pilot with Servier, a mid-cap French pharmaceutical company, is designed to validate this value proposition in a real drug discovery workflow. If successful, the venture will expand to Sanofi in Gentilly and to oncology and antiviral pipelines across European biotech hubs. SHORT ESSAY: ACTIVE BUSINESS DEVELOPMENT AND CUSTOMER TRACTION Active business development began in January 2025. I identified Servier as the ideal first pilot partner because of their oncology pipeline and their location in Suresnes, within the Ile-de-France biopharma cluster. I contacted the computational biology group at Servier directly with a technical brief showing the Platinum benchmark results and a proposal for a pilot study on a specific kinase target. A meeting was held in March 2025. The discussion is ongoing, with the next step being a data-sharing agreement for Servier to provide proprietary mutation panels for validation. Parallel to the Servier engagement, I established research partnerships with Paris-Saclay University, the Institute for Integrative Biology of the Cell, and Institut Pasteur. These partnerships provide access to biological validation capacity and to the SKEMPI 3K mutation dataset, which I will use to fine-tune the ESM-2 model and push the AUROC above 0.70 on the SKEMPI benchmark. The current SKEMPI 2.0 score of 0.634 is the baseline that fine-tuning will improve. Customer traction is pre-revenue but evidence-backed. The Platinum benchmark AUROC of 0.804 is a published, reproducible result. The model covers one hundred percent of mutations. The Servier pilot, if converted to a paid contract, will generate the first annual recurring revenue. I have also submitted applications to IncubAlliance and AI House in Paris, and I have a meeting scheduled with SEMIA and Quest for Health. The WILCO One BioTech programme in October 2026 is a confirmed pipeline target. RESEARCH STATEMENT The venture's core technology is a computational method for predicting drug resistance mutations from protein sequence alone. The method has three components. First, ESM-2, a 650-million-parameter protein language model developed by Meta AI, generates per-residue embeddings for the wild-type and mutant protein sequences. Second, the delta-embedding, the vector difference between the wild-type and mutant embeddings, captures the mutation's effect on the protein's learned evolutionary representation. Third, the delta-embedding is concatenated with the ECFP4 fingerprint of the drug molecule, and a Random Forest classifier predicts whether the mutation confers resistance to that drug. The model was trained and evaluated on the Platinum benchmark, a dataset of 553 mutations across 28 proteins with experimentally measured resistance labels. Protein-grouped cross-validation ensures that no mutations from the same protein appear in both training and test sets, which prevents data leakage. The AUROC of 0.804 plus or minus 0.025 exceeds the published state-of-the-art mCSM-lig AUROC of approximately 0.70. On the SKEMPI 2.0 benchmark, which includes binding affinity changes for protein-protein and protein-drug interactions, the model achieves an AUROC of 0.634. The next research milestone is fine-tuning ESM-2 on the SKEMPI 3K mutation dataset, which contains approximately 3,000 mutations with binding affinity labels. Fine-tuning will adapt the language model's representations to the specific task of resistance prediction, and I expect the SKEMPI 2.0 AUROC to rise above 0.70. This improvement is necessary to meet the validation threshold that Servier and other pharmaceutical partners require before signing pilot contracts. The long-term research agenda has three branches. First, extending the model to predict resistance for combination therapies, where two or more drugs are used simultaneously. Second, incorporating protein-protein interaction data to model resistance in the tumor microenvironment. Third, building a continuous learning pipeline that updates the model as new resistance mutations are reported in clinical literature and genomic databases. CHECKLIST - [ ] Motivation letter, 300 to 500 words, written in first person as Eniola Olutogun - [ ] Short essay on innovation in product, service, or market strategy, 200 to 350 words - [ ] Short essay on active business development and customer traction, 200 to 350 words - [ ] Research statement, 400 to 600 words - [ ] Proof of graduate degree (pharmacy or computational neuroscience transcripts) - [ ] Proof of Nigerian citizenship or permanent residence - [ ] Curriculum vitae or resume - [ ] Two letters of recommendation or professional references - [ ] Technical brief or one-page summary of Platinum benchmark results (AUROC 0.804) - [ ] Evidence of Servier pilot discussions (email correspondence or meeting notes) - [ ] Evidence of Paris-Saclay and Institut Pasteur partnerships (letters of support or agreements) - [ ] Completed fellowship application form on the Commonwealth Startup Fellowship website - [ ] Confirmation of availability for the full bootcamp schedule, 14 to 30 November 2025, in Accra EDITOR NOTES - Eligibility risk: The Commonwealth Startup Fellowship targets entrepreneurs from LMIC Commonwealth countries. Eniola is a Nigerian citizen, which qualifies. However, the fellowship may require that the startup be incorporated or in the process of incorporation. The venture is pre-incorporation. Confirm whether incorporation is a hard requirement or whether a detailed incorporation plan with a timeline suffices. - Fact verification needed: The Platinum benchmark AUROC of 0.804 is cited as published. Verify that this result is either published in a preprint or accepted manuscript, and include the DOI or preprint server link in the application. The fellowship may request independent verification of technical claims. - Gap in profile: The application materials do not specify Eniola's graduate degree details. The fellowship requires a graduate-level individual. Insert the specific degree title, institution, and year of graduation. If the degree is in pharmacy, confirm that it is a graduate-level degree (e.g., PharmD or MSc) and not an undergraduate BPharm. If the degree is in computational neuroscience, provide the institution and degree level.