MOTIVATION LETTER The DayOne Accelerator selects healthtech startups that solve pharmaceutical R&D bottlenecks. My platform, TOPOLOGIX, addresses one of the most expensive and unpredictable problems in drug development: predicting resistance mutations before clinical trials fail. Current tools like mCSM-lig achieve AUROC around 0.70 but cover only 18 percent of mutations because they require solved protein structures. TOPOLOGIX uses ESM-2 protein-language-model delta-embeddings combined with Morgan molecular fingerprints and a Random Forest classifier, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark of 553 mutations. It covers 100 percent of mutations because it works from sequence alone. I am an independent computational researcher based in Nigeria, currently enrolled in the M.Sc. Digital Health programme at Hasso Plattner Institute in Potsdam, Germany. My background combines a B.Pharm from the University of Ibadan with five years of software engineering and computational modeling experience. I have built and validated four independent research pipelines, published three sole-authored preprints on OSF and Zenodo, and have a co-authored paper under review at Alcohol. My collaborators include Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. The DayOne Accelerator is the right programme for TOPOLOGIX because it requires a prototype that can be offered to pharmaceutical companies within 12 to 18 months, and it demands a pharma-neutral approach. My platform is already functional: the Random Forest classifier runs on a DuckDB-based ingest pipeline, and I have tested it against two independent benchmarks. The next step is building a web-based interface where medicinal chemists can submit a drug candidate and receive a ranked list of likely resistance mutations within minutes. This directly serves the Accelerator's focus on AI drug discovery tools. Drug resistance is a global health priority with disproportionate impact on LMICs, including Nigeria. My perspective as a Nigerian pharmacist and researcher informs the platform's design: it must be accessible to groups that cannot afford structure-based tools or large compute clusters. TOPOLOGIX runs on a single GPU and uses open-source models throughout. I am applying to the DayOne Accelerator to transition TOPOLOGIX from a validated research prototype to a deployable product. The programme's network of pharmaceutical partners and its Basel location, at the center of European pharma, are the right environment for this transition. SHORT ESSAY: PROBLEM AND SOLUTION Pharmaceutical companies lose an estimated 30 percent of late-stage drug candidates to resistance mutations that were not predicted during early development. Current prediction tools require solved protein-ligand crystal structures, which exist for fewer than one in five clinically relevant mutation sites. This structural bottleneck means most resistance risks are discovered only during Phase II or Phase III trials, after millions of dollars have been spent. TOPOLOGIX eliminates the structural requirement. It uses ESM-2, a protein language model trained on 250 million sequences, to generate delta-embeddings that capture the effect of a mutation on the protein's representation space. These are combined with Morgan circular fingerprints of the drug molecule and fed into a Random Forest classifier. On the Platinum benchmark, the platform achieves AUROC 0.804, beating the best structure-based tool by 10 percentage points while covering five times more mutations. On SKEMPI 2.0, a more challenging benchmark of binding affinity changes, it achieves AUROC 0.634. The platform is designed for integration into existing pharma workflows. A medicinal chemist submits a SMILES string and a protein sequence; within minutes, the system returns a ranked list of single-point mutations most likely to confer resistance. The underlying pipeline is built on Python, DuckDB, and scipy, and can be deployed on a single Linux server with a GPU. SHORT ESSAY: AFRICA AND GLOBAL HEALTH Drug resistance is a present crisis for Africa. Multidrug-resistant tuberculosis accounts for 3.5 percent of new TB cases in Nigeria, and resistance to artemisinin-based combination therapies for malaria is documented across the continent. These diseases are neglected by commercial drug development because the market incentives are weak. When resistance emerges, there is no pipeline of backup compounds. TOPOLOGIX addresses this gap at the design stage. If a candidate drug for a neglected tropical disease can be screened against 10,000 possible resistance mutations before synthesis, the cost of failure drops and the probability of finding a durable compound rises. My platform makes this screening accessible to academic labs and small biotechs in LMICs because it requires no structural data and runs on modest hardware. My own trajectory reflects this priority. I trained as a pharmacist at the University of Ibadan, where I saw patients return with the same infections because first-line antibiotics had stopped working. That experience drove me to computational methods: I needed tools that could predict resistance before it appeared in the clinic. TOPOLOGIX is the result of that need. SHORT ESSAY: TECHNICAL APPROACH AND VALIDATION TOPOLOGIX uses a three-component architecture. First, ESM-2 generates per-residue embeddings for the wild-type and mutant protein sequences. The delta-embedding, the vector difference between the two, captures the mutation's effect on the protein's learned representation. Second, Morgan fingerprints with radius 2 and 2048 bits encode the drug molecule's topology. Third, a Random Forest classifier with 500 trees learns the mapping from the concatenated feature vector to a binary resistance label. I validated the platform against two benchmarks. On the Platinum dataset, which contains 553 mutations across 28 protein-drug complexes, the model achieved AUROC 0.804 with a standard deviation of 0.025 over five-fold cross-validation. On SKEMPI 2.0, which contains 4,168 mutations across 345 complexes, the model achieved AUROC 0.634. The drop reflects the greater diversity of the SKEMPI dataset, which includes non-resistance mutations and a wider range of binding affinity changes. I also tested whether simpler baselines could match this performance. A model using only Morgan fingerprints achieved AUROC 0.62 on Platinum. A model using only ESM-2 embeddings achieved 0.71. The combination is necessary for the observed gain. I pre-registered the analysis plan on OSF before running any experiments. SHORT ESSAY: COMMERCIAL PATHWAY The DayOne Accelerator requires a product that can be offered to pharmaceutical companies within 12 to 18 months. TOPOLOGIX meets this timeline. The core model is validated. The remaining work is building the user interface, the API layer, and the deployment infrastructure. Month 1 to 3: Build a web-based interface using React and a Python FastAPI backend. Users will upload a protein sequence and a drug SMILES string, and receive a ranked mutation list. Deploy on a cloud instance with GPU support. Month 4 to 6: Establish two pilot partnerships with mid-size pharma companies or biotechs. Offer free screening of up to 100 compounds per partner in exchange for feedback and validation data. Use the feedback to improve the model and the interface. Month 7 to 12: Launch a tiered pricing model. Academic labs and LMIC organizations pay a reduced rate. Commercial pharma companies pay per-compound or per-target fees. Target 10 paying clients by month 12. Month 13 to 18: Expand the platform to predict resistance combinations, not just single mutations. Add support for antibody drugs and CRISPR-based therapeutics. File a provisional patent on the delta-embedding method. The pharma-neutral requirement of the DayOne Accelerator is built into the platform's design. TOPOLOGIX does not favor any drug class or therapeutic area. It accepts any protein sequence and any small molecule, and returns predictions without bias toward any company's pipeline. CHECKLIST - [ ] Complete the DayOne Accelerator online application form at the programme website - [ ] Upload this motivation letter as a PDF - [ ] Upload the four short essays as a single PDF - [ ] Prepare a 3-minute pitch video demonstrating TOPOLOGIX on a live example - [ ] Gather two letters of reference: one from a research collaborator, one from a pharma or biotech contact - [ ] Prepare a one-page technical summary with the AUROC results and architecture diagram - [ ] Verify eligibility: confirm that TOPOLOGIX has received less than 10 million USD in dilutive funding - [ ] Confirm that the programme start date aligns with the M.Sc. Digital Health schedule at HPI EDITOR NOTES - Eligibility risk: The programme description says Pre-Seed to Series A with less than 10M in dilutive funding. Eniola has no funding at all, which is fine, but he should confirm that independent researchers with no company entity are eligible. If not, he may need to register a Nigerian or German GmbH before applying. - Fact verification: The 30 percent figure for late-stage drug candidate loss to resistance needs a citation. Eniola should find a published source or replace with a more conservative estimate. - Gap: The application asks for a prototype that can be offered to pharma within 12 to 18 months. Eniola should specify whether he has a co-founder or is applying solo. If solo, the timeline may need to be extended or a technical co-founder identified. - Gap: The essays assume Eniola will be in Germany during the programme. He should confirm that the DayOne Accelerator accepts remote participation or requires physical presence in Basel. If physical, he needs to arrange housing and visa. - Gap: The commercial pathway assumes Eniola can build a React frontend. His profile lists JavaScript and Node.js but not React specifically. He should either confirm his React skills or plan to hire a frontend developer.