MOTIVATION LETTER Drug-resistant infections kill nearly five million people each year, and Africa carries a disproportionate share of that burden. Standard resistance prediction tools require a protein structure, which exists for only about 18 percent of clinically relevant mutations. My TOPOLOGIX system predicts drug-resistance mutations from amino acid sequence alone, covering 100 percent of mutations with an AUROC of 0.804 plus or minus 0.025 on the Platinum benchmark, outperforming structure-based tools like mCSM-lig at roughly 0.70. This is the project I bring to the Africa Health-Tech Accelerator 2026 cohort. TOPOLOGIX combines ESM-2 protein language model delta-embeddings with Morgan/ECFP drug fingerprints and a Random Forest classifier. It requires no crystal structure, no homology model, no specialized laboratory equipment. A clinical lab in Lagos or Nairobi can run it on a standard laptop. The system currently predicts resistance for 553 mutations across the Platinum benchmark and generalizes to SKEMPI 2.0 at AUROC 0.634. The next step is a deployable web service that accepts a pathogen gene sequence and a candidate drug, then returns a resistance probability score within seconds. The accelerator's focus on healthcare innovation and trade matches this project's commercial path. African pharmaceutical manufacturers and diagnostic labs need resistance screening tools as antimicrobial stewardship programs expand across the continent. TOPOLOGIX is a productizable pipeline with a clear user, a clear workflow, and a measurable performance advantage over existing tools. My software engineering background supports the deployment plan: I have built four independent DuckDB-based ingest-to-analyze pipelines, self-hosted local LLM serving with llama.cpp, and operated production Linux systems with CI/CD and automated backup. I can take TOPOLOGIX from benchmark to beta. My path to this point is unconventional. I trained as a pharmacist at the University of Ibadan, worked in clinical and product roles, and built an independent computational research practice across neuroscience, protein ML, and dynamical systems. That independence produced a rigorous method: my cardiotoxicity topology study pre-registered a powered replication and reported honestly that topological features did not beat a plain descriptor baseline. TOPOLOGIX is the direct result of that negative result, a pivot from structure-based geometry to sequence-based representation. I am now enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute in Potsdam, which gives me access to European computational infrastructure while I remain rooted in African health challenges. The Africa Health-Tech Accelerator offers the structured pathway I need to move TOPOLOGIX from validated research to deployed product. I am applying for the 2026 cohort with a working prototype, a published benchmark record, and a clear deployment target. The commercial model is a software-as-a-service subscription for diagnostic labs and pharmaceutical companies, with a tiered pricing structure that keeps the tool accessible to public health institutions. I am ready to build the company, and I am ready to do it within the accelerator's framework. RESEARCH STATEMENT TOPOLOGIX is a sequence-based drug resistance prediction system built on protein language model embeddings. The problem it solves is structural: existing resistance prediction tools require a three-dimensional protein structure, but structural data covers only a fraction of clinically relevant mutations. The Platinum benchmark, a standard dataset of 553 resistance mutations, has structural coverage of roughly 18 percent. This means four out of five clinically important mutations cannot be scored by structure-based tools at all. TOPOLOGIX eliminates that dependency by learning resistance signals directly from amino acid sequence. The method works in three stages. First, an ESM-2 protein language model generates delta-embeddings, which capture the change in the protein's learned representation when a mutation is introduced. Second, the candidate drug is encoded as a Morgan/ECFP fingerprint, a standard topological representation of molecular structure. Third, a Random Forest classifier combines the protein delta-embedding with the drug fingerprint to predict whether the mutation confers resistance. The system achieves AUROC 0.804 plus or minus 0.025 on the Platinum benchmark and 0.634 on SKEMPI 2.0, a binding affinity dataset used as an external generalization check. It outperforms mCSM-lig, a widely used structure-based predictor, which scores approximately 0.70 on comparable tasks, while covering the full mutation space that mCSM-lig cannot access. The design of TOPOLOGIX was informed by a deliberate negative result. My earlier work tested whether bipartite persistent homology, a topological data analysis method, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication found that topological features did not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I then applied the same topological constructs to drug resistance prediction and found they carried almost no signal, AUROC 0.425 and 0.485 on the Platinum benchmark. These results ruled out interface geometry as the driver of resistance and motivated the shift to sequence representation. TOPOLOGIX is the product of that evidence-based pivot. The current validation status is honest and specific. TOPOLOGIX has passed benchmark evaluation on Platinum and SKEMPI 2.0. It has not yet been tested on prospective clinical isolates, and that is the next milestone. The deployment plan is a two-phase rollout. Phase one is a retrospective validation on curated African pathogen sequence databases, including Mycobacterium tuberculosis and Plasmodium falciparum resistance datasets, to establish performance on regionally relevant organisms. Phase two is a prospective pilot with one or two diagnostic laboratories in Nigeria, comparing TOPOLOGIX predictions against phenotypic resistance testing. The technical infrastructure for deployment already exists: the pipeline is implemented in Python with scikit-learn, runs on commodity hardware, and can be containerized and served as a REST API. The commercial model targets two customer segments. The first is diagnostic laboratories that need rapid resistance screening to guide treatment decisions. The second is pharmaceutical manufacturers and research organizations conducting drug development for infectious diseases, where early resistance profiling can inform candidate selection. Pricing will be subscription-based with tiered access, including a reduced-cost tier for public health institutions. The Africa Health-Tech Accelerator's emphasis on commercial viability and healthcare innovation aligns directly with this plan. TOPOLOGIX is a tool waiting for users, not a paper waiting for citation. ESSAY: INNOVATION AND NOVELTY TOPOLOGIX introduces a methodological shift in drug resistance prediction: it removes the structure requirement entirely. Every existing tool in this space, including mCSM-lig, DUET, and other structure-based predictors, needs a protein crystal structure or a reliable homology model. For the majority of clinically relevant mutations, that structure does not exist. TOPOLOGIX uses ESM-2 protein language model delta-embeddings to represent the mutation's effect on the protein's learned sequence context, combined with Morgan fingerprints for the drug. This is a different input representation that changes what can be predicted, not an incremental improvement on an existing method. The novelty is supported by evidence. On the Platinum benchmark, TOPOLOGIX achieves AUROC 0.804 plus or minus 0.025 while covering all 553 mutations. Structure-based mCSM-lig achieves approximately 0.70 but can only score the subset of mutations with available structures. The comparison demonstrates that sequence-based representation can outperform structure-based representation even on the structure-covered subset, while also handling the uncovered majority. The generalization to SKEMPI 2.0 at AUROC 0.634 shows the method transfers beyond its training distribution. The innovation also includes a methodological honesty that is rare in this field. My prior work on topological data analysis for cardiotoxicity and resistance prediction produced negative results, and I published them as negative results. The decision to abandon interface geometry and adopt sequence representation was driven by data, not by attachment to a method. That rigor is built into TOPOLOGIX's development process: every component is benchmarked against a plain baseline, and the system's claims are limited to what the benchmarks support. ESSAY: COMMERCIAL VIABILITY AND MARKET TRACTION The market for antimicrobial resistance prediction is growing as global health agencies push for better stewardship of existing drugs. The World Health Organization has identified antimicrobial resistance as one of the top ten global public health threats. African diagnostic laboratories and pharmaceutical companies are under increasing pressure to screen for resistance, but they lack access to tools that work without expensive structural biology infrastructure. TOPOLOGIX fills that gap with a software-only solution that runs on standard hardware. The commercial path is a software-as-a-service subscription. Target customers are diagnostic laboratories processing infectious disease samples, pharmaceutical companies developing drugs for African markets, and public health institutions running surveillance programs. The pricing model includes a standard tier for commercial labs, a premium tier for pharmaceutical R&D with API access and batch processing, and a reduced-cost tier for public health institutions. The unit economics are favorable: the marginal cost of serving an additional prediction is near zero, and the infrastructure requirements are modest. Traction to date is research validation rather than revenue. The system has demonstrated benchmark performance that beats existing tools on coverage and accuracy. The next milestone is a retrospective validation on African pathogen datasets, followed by a prospective pilot with diagnostic laboratories in Nigeria. The Africa Health-Tech Accelerator's network and mentorship can accelerate both the validation and the customer discovery process. I am presenting a validated technical core with a clear path to deployment and a defined customer base, not claiming product-market fit that does not exist. ESSAY: SCALABILITY AND IMPACT The scalability of TOPOLOGIX is inherent in its design. Because it requires only sequence data, it can be deployed anywhere with internet access or even offline on a local server. A diagnostic lab in a secondary hospital in Nigeria can run resistance predictions without investing in structural biology equipment or specialized bioinformatics staff. The system can be extended to new pathogens by fine-tuning the protein language model on new sequence data, and the Random Forest classifier can be retrained as new resistance data becomes available. The impact target is direct: faster and more accurate resistance detection means better treatment decisions, less use of ineffective drugs, and slower spread of resistant strains. In African health systems, where phenotypic resistance testing is often unavailable or delayed, a sequence-based predictor can provide same-day results. The potential scale is continental. The Africa Centres for Disease Control and Prevention has identified antimicrobial resistance surveillance as a priority, and TOPOLOGIX could serve as a standardized screening layer across multiple national surveillance programs. The team and execution capability are established. I have a B.Pharm from the University of Ibadan, clinical pharmacy experience, and a track record of independent computational research with reproducible pipelines and pre-registered protocols. I am currently enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute, which provides access to European computational research infrastructure. The technical skills required to build and deploy TOPOLOGIX, including Python, scikit-learn, Docker, and REST API development, are already in place. What the accelerator provides is the business development framework, the African health-tech network, and the structured path from research prototype to commercial product. CHECKLIST - [ ] Confirm the Africa Health-Tech Accelerator 2026 cohort application deadline from the programme website - [ ] Verify whether the accelerator requires a registered entity or accepts independent researchers for funding - [ ] Prepare a one-page executive summary of TOPOLOGIX with benchmark results and deployment plan - [ ] Prepare a technical appendix describing the ESM-2 delta-embedding method, Morgan fingerprint encoding, and Random Forest architecture - [ ] Prepare a slide deck (10-12 slides) covering problem, method, validation, market, and business model - [ ] Gather verification documents for the Platinum benchmark and SKEMPI 2.0 results (code repository, run logs) - [ ] Prepare a short video demo of the TOPOLOGIX pipeline running on a sample mutation - [ ] Identify two references who can speak to technical capability and research rigor - [ ] Draft a budget outline for the pilot phase (retrospective validation on African pathogen datasets and prospective lab pilot in Nigeria) - [ ] Confirm enrollment status at Hasso Plattner Institute and obtain current transcript - [ ] Prepare a one-page CV formatted for the accelerator's application portal - [ ] Verify ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) are current and linked to TOPOLOGIX code EDITOR NOTES - Research line selected: TOPOLOGIX. This is the only project in the profile with a productizable output, benchmark validation, and a direct commercial application for African health systems. The CCT model, neurocascade, ergofluids, and psyche-twin are research projects without a clear health-tech product path. The cardiotoxicity and resistance topology studies are negative results that motivated TOPOLOGIX and are referenced as such, not as current work. - Eligibility risk: The programme may require a registered entity for funding. The profile lists the applicant as an independent researcher with no registered company. The application should clarify that TOPOLOGIX can be spun out as a company during the accelerator, but this needs verification against the programme's terms. - Facts to verify: The AUROC values for TOPOLOGIX (0.804 plus or minus 0.025 on Platinum, 0.634 on SKEMPI 2.0) and the mCSM-lig comparison (approximately 0.70) must be checked against the actual benchmark outputs. The claim that structure-based tools cover only 18 percent of mutations needs a citation or a reproducible calculation from the Platinum dataset. - Gap to fill: The applicant must insert specific details about the Nigerian diagnostic laboratory pilot contacts, any existing conversations with potential customers, and the current state of the TOPOLOGIX codebase (is it a public repository with documentation, or a private prototype?). The application should also state whether the applicant has any prior business or entrepreneurship experience, as the accelerator will likely probe this.