MOTIVATION LETTER Antimicrobial resistance kills nearly five million people per year, and sub-Saharan Africa carries the highest burden per capita while possessing the least genomic surveillance infrastructure to track it. The SG-NAPI seed grant exists to change that imbalance by funding African researchers who build local capacity rather than exporting problems to external laboratories. My research programme does exactly that. I am a Nigerian pharmacist and computational researcher, currently enrolled in the M.Sc. Digital Health programme at the Hasso Plattner Institute in Potsdam, and I am applying to SG-NAPI to seed TOPOLOGIX, a sequence-based machine learning pipeline that predicts drug-resistance mutations from protein language model embeddings alone. TOPOLOGIX emerged from a falsified hypothesis. In 2024, I tested whether bipartite persistent homology could predict hERG cardiotoxicity from protein-ligand interface geometry. The pre-registered, powered replication found topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I then applied the same topological constructs to drug-resistance prediction on the Platinum benchmark and found they carry almost no signal, AUROC 0.425 and 0.485. Rather than abandon the question, I pivoted to sequence representations. TOPOLOGIX now uses ESM-2 delta-embeddings plus Morgan fingerprints and a Random Forest classifier, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark across 553 mutations, and 0.634 on SKEMPI 2.0. It beats structure-based baselines such as mCSM-lig at approximately 0.70 while covering 100 percent of mutations, versus roughly 18 percent for structure-limited tools. SG-NAPI is the correct home for this work because its selection criteria explicitly name AI, biotechnology, and medical sciences, and because its mission is to build research capacity in Africa. My prior experience with the Genomic Surveillance of Antimicrobial Resistance project at GHRU-GSAR involved building AMR genomics surveillance pipelines for Nigerian samples. TOPOLOGIX is the natural extension: a tool that can predict resistance mutations from sequence data alone, without requiring crystallographic structures that rarely exist for African clinical isolates. The seed grant would fund the first application of TOPOLOGIX to African pathogen genomic data, producing pilot results that demonstrate the method works on locally relevant organisms. I meet the eligibility conditions. I am a Nigerian national, 29 years old, and I hold a B.Pharm from the University of Ibadan with a German equivalent grade of 1.9. I am an independent researcher with a publication record that includes three sole-authored preprints under review at peer-reviewed journals and a co-authored paper in Alcohol under review at Elsevier. I have the technical infrastructure to execute this project with minimal resources: Python, public benchmarks, and my existing four DuckDB-based ingest-to-analyze pipelines. The grant would allow me to generate the pilot data needed to secure a longer-term position at a Nigerian research institution upon completion of my M.Sc. The problem is urgent, the method is validated, and the capacity-building angle is concrete. I am requesting SG-NAPI support to turn TOPOLOGIX from a benchmark success into a deployed tool for African antimicrobial resistance surveillance. RESEARCH STATEMENT TOPOLOGIX is a computational pipeline for predicting drug-resistance mutations from protein sequence alone. The core innovation is the use of ESM-2 protein language model delta-embeddings, which capture the functional impact of a mutation by comparing the embedding of the wild-type sequence to that of the mutant sequence. These delta-embeddings are combined with Morgan/ECFP drug fingerprints and fed into a Random Forest classifier. The method achieves AUROC 0.804 plus or minus 0.025 on the Platinum benchmark, which contains 553 mutations across diverse protein-drug systems, and AUROC 0.634 on SKEMPI 2.0. This outperforms structure-based tools such as mCSM-lig, which achieves approximately 0.70 but can only score mutations for proteins with known crystal structures, covering roughly 18 percent of the Platinum benchmark. TOPOLOGIX covers 100 percent of mutations because it requires only sequence data. The scientific motivation comes from a negative result I published as a pre-registered replication. In 2024, I tested whether bipartite persistent homology, an opposition-distance metric computed with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. The powered replication found topological features do 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 carry almost no signal, AUROC 0.425 and 0.485 on the Platinum benchmark. These results rule out interface geometry as the driver of resistance and motivate the sequence-representation approach that TOPOLOGIX now uses. The negative findings are published as preprints and are under review; they are not hidden. The proposed work for SG-NAPI has three objectives. First, retrain and validate TOPOLOGIX on African pathogen genomic datasets, specifically Mycobacterium tuberculosis and Salmonella Typhi sequences from Nigerian clinical isolates. I have prior experience with such data through my work at GHRU-GSAR, where I built AMR genomics surveillance pipelines. Second, benchmark TOPOLOGIX against existing resistance-prediction tools on these African datasets, quantifying the coverage advantage that sequence-based methods provide when structural data are absent. Third, publish the results as an open-access preprint and deposit the trained model and code on Zenodo and GitHub, so that other African researchers can apply the pipeline to their own pathogen datasets without requiring high-performance computing infrastructure. The methodology is feasible with minimal resources. ESM-2 embeddings can be generated on a single GPU or even CPU for small proteins. The Random Forest classifier trains in minutes. The Platinum and SKEMPI 2.0 benchmarks are publicly available. My existing infrastructure includes four DuckDB-based ingest-to-analyze pipelines and self-hosted local LLM serving via llama.cpp, which I will use for literature retrieval and annotation. The total compute requirement is modest and well within the budget of a seed grant. The expected outcomes are concrete. A validated model for African pathogen resistance prediction, measured by AUROC and coverage on held-out African genomic data. A public benchmark dataset of Nigerian clinical isolate sequences with resistance labels, curated from GHRU-GSAR and public repositories. An open-access publication describing the method and its performance. A trained model artifact that other researchers can download and apply to their own data. These outcomes directly serve SG-NAPI's goal of building research capacity in Africa, because they produce reusable tools and data rather than a single publication. The long-term vision is a fully deployed resistance-prediction service for African clinical microbiology laboratories, accessible via a web interface, that accepts a pathogen genome sequence and returns predicted resistance mutations with confidence scores. The SG-NAPI seed grant funds the first two years of that vision: the validation on African data and the public release of the model. Subsequent funding from larger grants, which I will apply for with the pilot data this grant produces, will support the deployment and maintenance of the service. ESSAY RESPONSE: AFRICA ANGLE AND CAPACITY BUILDING Antimicrobial resistance surveillance in Africa is limited by infrastructure, not by biology. Whole-genome sequencing is increasingly affordable, but the analysis pipelines that interpret those genomes are developed in Europe and North America, trained on European and North American isolate collections, and validated on proteins with known crystal structures. African clinical isolates are underrepresented in those training sets, and the resistance mutations that matter locally are often missed because they are not in the training data. TOPOLOGIX addresses this directly because it requires only sequence data, which African laboratories can generate, and it can be retrained on local isolate collections without requiring structural biology expertise. My prior work at GHRU-GSAR built AMR genomics surveillance pipelines for Nigerian samples, and I saw firsthand that the bottleneck is not sequencing capacity but analysis capacity. The SG-NAPI seed grant would fund the first systematic application of a sequence-based resistance predictor to Nigerian pathogen genomic data, producing a public benchmark dataset and a trained model that other African researchers can use. The capacity-building component is explicit: I will publish the code, the model, and the data on open repositories, and I will document the pipeline so that a researcher with Python proficiency can reproduce the entire workflow. This is a tool built by an African researcher, trained on African data, and released for African use. ESSAY RESPONSE: FEASIBILITY AND PILOT DATA The project is designed to produce pilot data within six months of funding. The Platinum and SKEMPI 2.0 benchmarks are already downloaded and processed. The ESM-2 embedding pipeline is already implemented and tested. The Random Forest classifier is already trained and validated on the Platinum benchmark. The remaining work is data acquisition and curation: obtaining Nigerian clinical isolate sequences with resistance labels from GHRU-GSAR and public repositories, formatting them for the existing pipeline, and running the validation. I estimate this requires two months of part-time work. The risk is low because every component has already been demonstrated to work on the benchmark data. The only external dependency is access to the African genomic datasets, which I have already used in my GHRU-GSAR role and can access through existing collaborations. CHECKLIST - [ ] Confirm SG-NAPI eligibility criteria on the programme website, specifically the requirement to hold or be securing a position at a local university or research institution in Nigeria - [ ] Verify that the M.Sc. enrolment at Hasso Plattner Institute does not conflict with the "returned to home country" or "newly appointed PI" eligibility conditions - [ ] Obtain a letter of institutional support from a Nigerian university or research institution, or document the plan to secure such a position - [ ] Prepare a detailed budget for the seed grant, itemizing compute costs, data access fees, and publication charges - [ ] Update ORCID record with all preprints and the co-authored Alcohol paper to ensure the publication list is current - [ ] Prepare a one-page CV formatted for the SG-NAPI application portal - [ ] Draft a data management plan covering the deposition of code, model artifacts, and curated African genomic datasets on Zenodo and GitHub - [ ] Identify and contact a collaborator at a Nigerian institution who can provide clinical isolate sequences and resistance labels - [ ] Submit the application before the deadline listed on the SG-NAPI programme website EDITOR NOTES - Eligibility risk: the SG-NAPI criteria mention a PhD obtained abroad within the last five years and a return to the home country. Eniola does not hold a PhD and is currently enrolled in an M.Sc. in Germany. This is a significant mismatch that must be verified against the actual programme terms before submission. If the PhD requirement is strict, this application should not proceed. - The institutional affiliation requirement is unmet. Eniola is listed as an independent researcher. A letter of support from a Nigerian university or research institution is required, and the profile does not indicate an existing affiliation. This must be secured before submission. - The TOPOLOGIX results on Platinum and SKEMPI 2.0 are from the applicant's own benchmarks and have not yet been peer-reviewed. The application should present these as internal validation results, not as published findings, and should note that the negative topology results are under review at peer-reviewed journals.