MOTIVATION LETTER Antimicrobial resistance kills over one million people annually. Existing drug-resistance prediction tools require protein structures, covering only 18 percent of known mutations. My software platform, TOPOLOGIX, solves this by predicting resistance mutations from protein sequence alone, using ESM-2 protein-language-model embeddings and Morgan fingerprints with a Random Forest classifier. On the Platinum benchmark of 553 mutations, TOPOLOGIX achieves an AUROC of 0.804 plus or minus 0.025, beating structure-based baselines like mCSM-lig at approximately 0.70 while covering 100 percent of mutations. This is a direct commercial opportunity for the Innovate UK Biomedical Catalyst Accelerator. I am an independent computational researcher, a licensed pharmacist, and an incoming M.Sc. Digital Health student at the Hasso Plattner Institute in Germany. My research spans addiction neuroscience, protein machine learning, and dynamical-systems methods. I have sole-authored three preprints on the Conjunctive Consolidation Threshold model of reward-memory encoding, co-authored a paper currently under review at Alcohol, and built a receptor-to-behavior brain-circuit simulation engine called neurocascade with 62 passing tests. My work on hERG cardiotoxicity topology, published as a pre-registered replication, demonstrated that topological features do not beat a plain descriptor baseline, settling a question the literature had never properly tested. TOPOLOGIX targets the antimicrobial resistance priority area of this accelerator. The platform is at TRL 3, validated on two independent benchmarks. The commercial route is a software-as-a-service licensing model to pharmaceutical companies and contract research organizations conducting preclinical drug development. A UK-based SME can be formed to host the platform, with development costs in the range of 50,000 to 100,000 pounds for a minimum viable product with a web interface and API. My Nigerian background provides direct insight into regions where AMR burden is highest, strengthening the global health case. The Pre-Accelerator phase in November to December 2025 is the correct entry point for this technology. I am prepared to commit full-time effort during the Accelerator phase from January to March 2026. The programme's focus on early-stage technologies at TRL 2 to 4 matches TOPOLOGIX exactly. I seek funding to refine the model, build a demonstration interface, and conduct validation on a third independent dataset from the SKEMPI 2.0 benchmark, where current performance is 0.634 AUROC, indicating room for improvement through additional feature engineering. RESEARCH STATEMENT TOPOLOGIX addresses a specific gap in computational drug development: predicting whether a mutation in a target protein will confer resistance to a candidate drug. Current tools require a protein-ligand crystal structure, which exists for fewer than one in five clinically observed mutations. This structural bottleneck means resistance screening is incomplete during early drug discovery, leading to late-stage failures and contributing to the antimicrobial resistance crisis. My approach replaces structure with sequence representation. I use 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 learned representation space. These are concatenated with Morgan circular fingerprints of the drug molecule and passed to a Random Forest classifier. The method requires only the wild-type sequence, the mutant sequence, and the drug's SMILES string. No docking, no homology modeling, no crystal structure. Results on the Platinum benchmark, a curated set of 553 resistance-associated mutations across 10 drug-target pairs, show an AUROC of 0.804 with a standard deviation of 0.025 over cross-validation folds. This exceeds the best structure-based method, mCSM-lig, at approximately 0.70. On the SKEMPI 2.0 benchmark of binding free energy changes, the AUROC is 0.634, indicating the method generalizes to related tasks but requires further optimization for binding affinity prediction. The next development steps are threefold. First, incorporate attention-based pooling of ESM-2 embeddings to replace the current mean-pooling, which may improve signal extraction. Second, train on the full Platinum dataset with a held-out test set of 20 percent to establish a rigorous generalization benchmark. Third, build a web-based demonstration interface using a Python backend and a JavaScript frontend, allowing users to input a protein sequence and drug SMILES and receive a resistance probability within seconds. The commercial model is straightforward. Pharmaceutical companies and CROs pay a per-mutation or per-target subscription fee for access to the prediction API. A tiered pricing structure based on volume, from 0.50 dollars per mutation for academic users to 0.10 dollars per mutation for enterprise contracts, would generate revenue at scale. The total addressable market includes every preclinical drug development program targeting infectious diseases and oncology, where resistance mutations are a primary failure mode. My independent researcher status means I have built TOPOLOGIX without institutional overhead, using open-source tools and self-hosted compute. The platform runs on a Linux VPS with Caddy TLS and automated backup, using DuckDB for data management and llama.cpp for local LLM serving. This lean infrastructure keeps operating costs below 100 dollars per month. The Innovate UK Biomedical Catalyst Accelerator would provide the resources to professionalize the platform, conduct formal usability testing with pharmaceutical partners, and establish a UK-registered company. PROJECT PLAN Phase 1, months 1 to 3: Model refinement. Implement attention-based ESM-2 embedding pooling. Train on 80 percent of Platinum benchmark, test on held-out 20 percent. Target AUROC of 0.85 or higher. Deliverable: updated model weights and a technical report. Phase 2, months 4 to 6: Interface development. Build a web application with a Python backend using FastAPI and a JavaScript frontend using React. Deploy on a UK-based cloud server. Implement user authentication, job queue, and result storage. Deliverable: functional web interface with API documentation. Phase 3, months 7 to 9: Validation and partnership. Test on a third independent dataset, the DREAM challenge resistance prediction data. Contact two UK-based pharmaceutical companies or CROs for beta testing. Collect user feedback and performance metrics. Deliverable: validation report and at least one letter of interest from a potential customer. Phase 4, months 10 to 12: Commercial launch. Register a UK private limited company. Set up payment processing through Stripe. Launch public API with free tier for academic users and paid tier for commercial users. Deliverable: live service with paying customers. Budget: 60,000 pounds total. Cloud compute and storage, 12,000 pounds. Software development contractor for frontend, 18,000 pounds. Legal and company registration, 3,000 pounds. Travel for partnership meetings, 5,000 pounds. My stipend for 12 months at 1,500 pounds per month, 18,000 pounds. Contingency, 4,000 pounds. CHECKLIST - [ ] Complete online application form on Innovate UK Business Connect website - [ ] Upload motivation letter as PDF, 500 words maximum - [ ] Upload research statement as PDF, 600 words maximum - [ ] Upload project plan as PDF, 400 words maximum - [ ] Provide evidence of eligibility: UK-based SME registration or partnership letter from UK academic institution - [ ] Confirm TRL level: TRL 3, with evidence from benchmark results - [ ] Prepare a 3-minute video pitch explaining TOPOLOGIX and its commercial potential - [ ] Submit by the programme deadline, check website for exact date EDITOR NOTES - Eligibility risk: The programme requires a UK-based SME, spinout, or academic. Eniola is an independent researcher based in Nigeria and Germany. He must either partner with a UK academic institution or register a UK company before applying. The project plan assumes company registration in month 10, but the application requires SME status at submission. This needs resolution before applying. - The budget line for Eniola's stipend at 1,500 pounds per month may be below UK minimum wage for a full-time role. Check current UK National Living Wage rates and adjust the budget accordingly, or frame the stipend as a partial living allowance with the remainder covered by other sources. - The SKEMPI 2.0 AUROC of 0.634 is significantly lower than the Platinum benchmark. The application should acknowledge this gap and explain why it represents an opportunity for improvement rather than a weakness. The project plan's Phase 1 target of 0.85 AUROC on Platinum is ambitious; verify that the attention-based pooling approach has literature support for this magnitude of improvement. - Eniola's M.Sc. Digital Health at HPI starts in Winter Semester 2026/27, which overlaps with the Accelerator phase from January to March 2026. Confirm that the programme schedule does not conflict with coursework or that the university permits part-time study alongside commercial activity. - The application mentions a 3-minute video pitch as a requirement. Verify this on the programme website, as the captured data does not specify video requirements. If required, prepare a script that demonstrates the software interface and walks through a single prediction example.