MOTIVATION LETTER The $500K investment and the YC network are the two resources my work needs most. TOPOLOGIX is a protein-language-model platform that predicts drug-resistance mutations from sequence alone. It covers 100 percent of mutations in the Platinum benchmark, 553 total, while structure-based tools like mCSM-lig cover roughly 18 percent. My model achieves an AUROC of 0.804 plus or minus 0.025 on that benchmark, beating mCSM-lig's approximate 0.70. This is a commercial product with a clear buyer: pharmaceutical companies losing billions to resistance failures in late-stage trials. I am a pharmacist, a computational modeler, and a software engineer. I hold a B.Pharm from the University of Ibadan with a German-equivalent grade of 1.9. I built TOPOLOGIX as an independent researcher, which means I shipped the entire stack myself: ESM-2 delta-embeddings, Morgan fingerprints, the Random Forest classifier, and the evaluation pipeline. I also built four separate DuckDB-based data pipelines and self-hosted LLM serving infrastructure, so I can run the full product lifecycle without a team. YC funds velocity, not decks. My velocity is documented across six preprints and three journal submissions in the last two years. The market is large and growing. Antimicrobial resistance is projected to kill 10 million people per year by 2050. Cancer drug resistance accounts for over 90 percent of deaths in metastatic cancers. Every pharmaceutical company with a small-molecule pipeline needs resistance prediction before trials, not after. TOPOLOGIX sells as an API or SaaS tier, priced per mutation screened, with an enterprise tier for internal deployment behind firewalls. YC's Request for Startups explicitly names AI for biotech and drug development. TOPOLOGIX fits that mandate directly. The YC S26 batch requires relocation to San Francisco, which I am prepared to do. The batch structure, the partner network, and the demo day exposure are the fastest path to pharmaceutical partnerships and follow-on capital. I am applying with a working model, a falsified null result that shaped its design, and a benchmark that beats the published baseline. The next step is a pilot with a pharma partner and a web interface for self-serve screening. YC is the right accelerator to make that happen. RESEARCH STATEMENT TOPOLOGIX predicts drug-resistance mutations from protein sequence alone, using protein-language-model embeddings and drug fingerprints. The system addresses a specific failure in the field: structure-based tools like mCSM-lig require a resolved crystal structure, which exists for only about 18 percent of clinically relevant mutations. My model needs only the sequence, so it covers the full 553-mutation Platinum benchmark. The technical core is a delta-embedding approach. I feed the wild-type and mutant protein sequences through ESM-2, a 650-million-parameter protein language model, and take the difference between their embeddings. That delta vector captures the mutational perturbation in a learned representation space. I concatenate it with a Morgan/ECFP fingerprint of the drug molecule and train a Random Forest classifier. The result is an AUROC of 0.804 plus or minus 0.025 on Platinum, and 0.634 on SKEMPI 2.0. The structure-based baseline, mCSM-lig, scores approximately 0.70 on the subset it can actually process. The design was shaped by a falsified hypothesis. My earlier work tested whether bipartite persistent homology of protein-ligand interfaces could predict resistance. It could not. On the Platinum benchmark, the topological features scored AUROC 0.425 and 0.485, no better than chance. I pre-registered that study, ran it with adequate power, and published the null result. That negative finding ruled out interface geometry as the driver of resistance and motivated the sequence-representation approach that became TOPOLOGIX. The null result is the reason the current model works. The commercial application is straightforward. Pharmaceutical companies screen candidate drugs against known resistance mutations before committing to clinical trials. Current tools miss most mutations because they need structures. TOPOLOGIX screens all of them from sequence. The product is an API that accepts a protein sequence and a drug SMILES string, and returns a resistance probability per mutation. Pricing is per mutation screened, with volume discounts and an enterprise tier for on-premise deployment. The validation status is honest. The model performs well on retrospective benchmarks. It has not yet been validated prospectively against a live drug-development program. That is the purpose of the pilot phase. I am seeking a pharmaceutical partner with a resistance problem in their pipeline, and I will publish the prospective results regardless of outcome. My background supports this work directly. I am a licensed pharmacist, so I understand the biology of resistance mechanisms. I built the entire software stack myself, from the ESM-2 inference pipeline to the evaluation harness. I have published six preprints and have three sole-authored papers under review at peer-reviewed journals. The next milestone is a self-serve web interface and a pilot with a pharma partner, both achievable within the YC batch timeline. ESSAY: WHY THIS FOUNDER The strongest evidence that I can execute is the trajectory from a failed hypothesis to a working product. In 2024, I tested whether topological data analysis of protein-ligand interfaces could predict drug resistance. The pre-registered, powered replication found no signal: AUROC 0.425 and 0.485 on the Platinum benchmark. I reported the null result directly rather than reframing it. That discipline is rare in computational biology, where negative results often disappear. The null result redirected the work. If interface geometry did not drive resistance, the signal had to be in the sequence. I pivoted to protein-language-model embeddings, built the delta-embedding pipeline, and achieved an AUROC of 0.804 on the same benchmark. The model covers 100 percent of mutations versus 18 percent for structure-based tools. That is a product. I have shipped consistently as an independent researcher. Six preprints in two years. Three sole-authored papers under review. A co-authored paper under review at Alcohol, an Elsevier journal. A brain-circuit simulation engine with 62 passing tests. Four separate data pipelines. Self-hosted LLM serving infrastructure. I do not need a team to build; I need a team to scale. The YC network is the specific resource I lack. I have endorsements from Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. Those relationships validate my research. They do not provide pharmaceutical partnerships, pricing guidance, or Series A introductions. YC does. I am 29, Nigerian, and enrolled in the M.Sc. Digital Health program at Hasso Plattner Institute in Potsdam, starting Winter 2026/27. I am prepared to relocate to San Francisco for the batch. I am applying as a solo founder because the work is solo by construction, but I understand YC's scrutiny of solo founders and I have the shipping record to answer it. ESSAY: MARKET AND TRACTION The market for drug-resistance prediction is defined by two numbers. Antimicrobial resistance is projected to cause 10 million deaths per year by 2050. Cancer drug resistance accounts for over 90 percent of deaths in metastatic cancers. Every pharmaceutical company with a small-molecule pipeline needs to know, before trials, whether a candidate drug will be defeated by known resistance mutations. The current tools fail at this task. Structure-based predictors like mCSM-lig require resolved crystal structures, which exist for only about 18 percent of clinically relevant mutations. The remaining 82 percent are invisible to them. TOPOLOGIX requires only the protein sequence, so it screens the full set. My traction is the benchmark itself. On the Platinum benchmark of 553 mutations, TOPOLOGIX achieves AUROC 0.804 plus or minus 0.025. The structure-based baseline scores approximately 0.70 on the subset it can process. On SKEMPI 2.0, the model scores 0.634. These are published, reproducible numbers, not deck claims. The commercial model is an API. A pharmaceutical company submits a protein sequence and a drug SMILES string. The API returns a resistance probability per mutation. Pricing is per mutation screened, with an enterprise tier for on-premise deployment behind the company's firewall. The target customers are the top 50 pharmaceutical companies by R&D spend, plus the growing number of biotech firms with resistance problems in their pipelines. The honest limitation is that the model has not yet been validated prospectively. Retrospective benchmarks are necessary but not sufficient. The pilot phase will pair TOPOLOGIX with a pharmaceutical partner's live resistance problem, and I will publish the prospective results whether they confirm or refute the model. That commitment to publishing negative results is what separates this product from the typical accelerator pitch. The YC S26 batch provides the specific resources this venture needs: pharmaceutical network access, pricing and packaging guidance, and the demo day platform for Series A fundraising. The $500K investment funds the pilot phase and the web interface. The relocation to San Francisco is a requirement I accept. CHECKLIST - [ ] Complete YC S26 online application at https://ycinsight.com/yc-s26 - [ ] Submit one-minute founder video, filmed in landscape, speaking directly to camera - [ ] Verify all benchmark numbers against the Platinum and SKEMPI 2.0 evaluation scripts before submission - [ ] Confirm the exact wording of the mCSM-lig baseline comparison (0.70 approximate) with the published source - [ ] Prepare a one-page technical appendix describing the ESM-2 delta-embedding architecture for the application's technical questions - [ ] Draft a response to the question about co-founders, stating solo founder status and the shipping record that justifies it - [ ] Confirm relocation timeline to San Francisco for the S26 batch dates - [ ] Prepare a list of three pharmaceutical companies as target pilot partners, with named contacts where possible - [ ] Verify the current status of the three sole-authored preprints under review and list them accurately in the application - [ ] Confirm the M.Sc. Digital Health enrollment at HPI/Potsdam does not conflict with the batch schedule EDITOR NOTES - Eligibility risk: YC requires founders to relocate to San Francisco for the batch. Eniola is enrolled in an M.Sc. program at HPI/Potsdam starting Winter 2026/27. The batch and the semester may overlap. This needs a concrete plan before submission, not a vague commitment. - The mCSM-lig baseline of approximately 0.70 is cited from memory of the literature. The exact number and the benchmark subset it applies to must be verified against the published paper before this goes into the application. YC partners will check. - The Platinum benchmark AUROC of 0.804 plus or minus 0.025 and the SKEMPI 2.0 score of 0.634 must be reproducible from the evaluation scripts. Run the full pipeline once more and attach the output to the application. - The application should not mention the CCT model, neurocascade, ergofluids, or psyche-twin. They are not relevant to YC's selection criteria and will dilute the TOPOLOGIX pitch. If asked about other work, mention the falsified topology result as the design driver, nothing else. - The essay on market and traction claims a pilot phase with a pharmaceutical partner. No partner is named. Eniola must either name a real contact or soften the claim to "targeting a pilot with a pharmaceutical partner" before submission.