MOTIVATION LETTER The Commonwealth Startup Fellowship targets entrepreneurs from low and middle-income Commonwealth countries who can build ventures that drive economic growth and solve pressing challenges. My venture predicts drug resistance mutations from protein sequence alone, without requiring crystal structures. This capability directly addresses antimicrobial resistance and oncology treatment failure, two crises that disproportionately affect countries like Nigeria where I was born and trained as a pharmacist. I am Eniola Olutogun, a pharmacist turned machine learning engineer. My venture uses ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints, classified by a Random Forest model. On the Platinum benchmark of 357 mutations, the system achieves an AUROC of 0.634 using protein-grouped cross-validation. Critically, it covers 100 percent of mutations, compared to 17.6 percent for structure-limited tools. Published state-of-the-art mCSM-lig reports an AUROC of 0.70, so my system is within striking distance while requiring no crystallography data. This technology democratises drug discovery. Laboratories in resource-limited settings cannot routinely solve protein crystal structures. My platform removes that barrier. A researcher in Lagos or Nairobi can input a protein sequence and a drug molecule, and receive a prediction of which mutations will cause resistance, all without expensive equipment or structural biology expertise. The venture is pre-seed, proof-of-concept validated, and not yet incorporated. I am sole author and founder, operating from Ile-de-France with the intention to establish the company in the European Union. Named partners include Servier in Suresnes, Paris-Saclay through I2BC and Institut Pasteur, and Sanofi in Gentilly. A meeting with SEMIA and Quest for Health is in progress. Applications to IncubAlliance and AI House have been submitted. WILCO One BioTech is scheduled for October 2026. The Commonwealth Startup Fellowship is the right programme because it seeks entrepreneurs who will create scalable impact in their home countries and across the Commonwealth. My venture reduces drug development costs, improves treatment outcomes, and can be deployed directly in health systems that currently lack predictive tools for resistance. I am not applying to relocate permanently. I am applying to build a company that serves Commonwealth populations. The fellowship would fund the next milestone: fine-tuning ESM-2 on the SKEMPI 3K mutation dataset to push AUROC to 0.70 or above, then initiating a pilot with Servier. That pilot is the path to annual recurring revenue and a funded Series A. SHORT ESSAY: VENTURE OVERVIEW AND IMPACT Drug resistance kills over one million people per year globally. In oncology, acquired resistance to targeted therapies is the primary cause of treatment failure. Current computational tools for predicting resistance mutations require solved protein crystal structures, which exist for fewer than 20 percent of clinically relevant proteins. This structural bottleneck excludes the majority of drug-target pairs from predictive analysis. My venture eliminates that bottleneck. The core technology uses ESM-2, a protein language model trained on 250 million sequences, to generate delta-embeddings that capture mutation effects without any structural input. These embeddings are combined with ECFP4 drug fingerprints and classified by a Random Forest model. On the Platinum benchmark, the system achieves an AUROC of 0.634 with 100 percent mutation coverage. The comparable structure-dependent tool mCSM-lig achieves an AUROC of 0.70 but covers only 17.6 percent of mutations. The business model is software-as-a-service for pharmaceutical R&D teams. Customers pay per target or per compound library screened. The target market includes the 200-plus biotech and pharma companies actively developing kinase inhibitors, antibiotics, and antiviral therapies. Initial go-to-market strategy focuses on European mid-cap pharma, starting with Servier as a pilot partner. Impact is measurable in two dimensions. First, reducing the cost of drug development by enabling earlier identification of resistance liabilities. Second, improving patient outcomes by informing combination therapy design and treatment sequencing. For Commonwealth countries with limited genomic surveillance infrastructure, the platform offers a low-cost, sequence-only method to anticipate resistance before it emerges in the clinic. SHORT ESSAY: FOUNDER BACKGROUND AND MOTIVATION I trained as a pharmacist at the University of Ibadan in Nigeria, where I saw first-hand how drug resistance turns treatable infections into fatal ones. Antibiotics that cost a few dollars per course become useless when resistance emerges, and second-line therapies are often unavailable or unaffordable. That experience drove me to understand resistance at the molecular level. After pharmacy school, I transitioned into computational neuroscience and machine learning. I developed the Conjunctive Consolidation Threshold model of memory consolidation, published preprints on computational pharmacology, and built platforms including IMPRINT and TOPOLOGIX for analysing neural and molecular data. My ORCID record documents this trajectory from pharmacology to computational research. The shift from academic research to entrepreneurship came from a simple observation. My prediction models worked in silico, but they were not being used by clinicians or drug developers. To create real-world impact, I needed to build a product, not just publish a paper. That is why I moved to Ile-de-France, enrolled in the AI and biotech ecosystem there, and began the process of incorporating a company. The Commonwealth Startup Fellowship matters to me because it connects my Nigerian origin with my European base. I am not leaving Nigeria behind. I am building a tool that can be deployed in Nigerian hospitals, Indian clinics, and Kenyan research labs. The fellowship would give me the entrepreneurial training, network, and credibility to make that deployment happen. RESEARCH STATEMENT My research addresses a fundamental gap in computational drug discovery: predicting drug resistance mutations without requiring protein crystal structures. The standard approach in the field uses structure-based features such as residue depth, solvent accessibility, and interatomic contacts. These features are powerful but unavailable for the majority of protein targets. My work replaces structural features with representations learned by protein language models, specifically ESM-2. The technical approach has three components. First, ESM-2 generates per-residue embeddings for wild-type and mutant protein sequences. The delta-embedding, the vector difference between mutant and wild-type at the mutation site, captures the local effect of the amino acid substitution. Second, the drug molecule is encoded as an ECFP4 fingerprint, a 2048-bit binary vector representing circular substructures. Third, the concatenated delta-embedding and fingerprint are classified by a Random Forest model trained on known resistance mutations. Validation was performed on the Platinum benchmark, a curated set of 357 mutations across multiple protein-drug pairs. Protein-grouped cross-validation ensures that no mutations from the same protein appear in both training and test sets. The model achieves an AUROC of 0.634 with 100 percent mutation coverage. For comparison, mCSM-lig, the current structure-dependent state-of-the-art, achieves an AUROC of 0.70 but covers only 17.6 percent of Platinum mutations because it requires a crystal structure. The next research milestone is fine-tuning ESM-2 on the SKEMPI 3K dataset, which contains over 3,000 mutation effects on protein-protein and protein-ligand binding. Fine-tuning will adapt the language model to the specific task of predicting binding affinity changes caused by mutations. I expect this to raise AUROC to 0.70 or above, matching structure-dependent methods while maintaining 100 percent coverage. Beyond the benchmark, the research programme includes three parallel tracks. First, expanding the mutation dataset by mining clinical resistance databases from oncology and antimicrobial resistance literature. Second, incorporating attention-based interpretability to identify which residues and drug substructures drive the resistance prediction. Third, building a web-based interface that allows a user to paste a protein sequence, draw or upload a drug molecule, and receive a resistance prediction in under one minute. The broader scientific contribution is demonstrating that protein language models can substitute for structural biology in a high-stakes prediction task. If a model trained only on sequences can match or exceed structure-based methods, then the field of computational drug discovery must reconsider its dependence on crystallography. This has implications for target classes that are difficult to crystallise, including membrane proteins, intrinsically disordered proteins, and proteins from pathogens that are hard to express in recombinant systems. My publication record includes preprints on computational pharmacology and the Conjunctive Consolidation Threshold model. I have not yet published the Platinum benchmark results in a peer-reviewed journal, but I intend to submit a manuscript to Bioinformatics or PLOS Computational Biology within six months. The fellowship would support the computational resources and protected time needed to complete that submission while simultaneously building the commercial product. CHECKLIST - [ ] Confirm eligibility: Commonwealth citizen from a low or middle-income Commonwealth country (Nigeria qualifies) - [ ] Verify fellowship start date and duration on programme website - [ ] Prepare proof of concept validation data (AUROC 0.634, Platinum benchmark results) - [ ] Draft one-page executive summary of the venture - [ ] Obtain letter of support from Servier or Paris-Saclay collaborator - [ ] Prepare financial projection for pre-seed to Series A runway - [ ] Confirm incorporation timeline for EU entity - [ ] Gather academic transcripts and pharmacy license from University of Ibadan - [ ] Update ORCID record with all preprints and platform publications - [ ] Prepare video pitch or slide deck if required by programme EDITOR NOTES - Eligibility risk: The programme website URL points to opportunitiesjournal.com, not an official Commonwealth body. Verify the actual provider and confirm that Nigerian citizens are eligible. Some Commonwealth programmes exclude Nigeria due to its upper-middle-income reclassification by the World Bank. - Fact to verify: The Platinum benchmark AUROC of 0.634 is reported as protein-grouped cross-validation. Confirm whether the programme expects a different validation metric or a comparison to additional baselines beyond mCSM-lig. - Gap: The profile does not specify whether Eniola has a co-founder or any team members. If the fellowship requires a team, this must be addressed. If sole founder is acceptable, state that explicitly. - Gap: No mention of revenue, paying customers, or letters of intent. The Servier pilot is described as future. If the programme values commercial traction, Eniola should insert any existing conversations, LOIs, or pilot agreements. - Gap: The profile lists named partners but does not specify whether any have committed funding, compute resources, or in-kind support. Clarify the nature of each relationship before submission.