MOTIVATION LETTER The NGI Zero Commons Fund exists to fund software that serves the public, not shareholders. TOPOLOGIX is exactly that kind of software: an open-source tool that predicts drug-resistance mutations from protein sequence alone, built to remove the structural-data bottleneck that excludes most of the world from resistance-guided drug design. This project is the direct result of a falsified hypothesis. My earlier work tested whether interface topology, measured with bipartite persistent homology, could predict drug resistance. It could not. On the Platinum benchmark, topological descriptors scored AUROC 0.425 and 0.485, below chance. I reported that result publicly rather than burying it, and then I built the tool that the negative result pointed to. TOPOLOGIX uses ESM-2 protein language model delta-embeddings combined with Morgan/ECFP drug fingerprints and a Random Forest classifier. It achieves AUROC 0.804 plus or minus 0.025 on the Platinum benchmark across 553 mutations, and 0.634 on SKEMPI 2.0. It beats the structure-based baseline mCSM-lig, which scores around 0.70, while covering 100 percent of mutations. Structure-limited tools like mCSM-lig cover roughly 18 percent of mutations because they require a resolved crystal structure. Most resistance-relevant proteins, especially from pathogens prevalent in Africa, have no such structure. TOPOLOGIX does not need one. That is the equity argument in one number: 100 percent coverage versus 18 percent. The NGI Commons Fund's stated values are openness, interoperability, decentralization, and trustworthiness. TOPOLOGIX is MIT-licensed, on GitHub at github.com/AmunRaPtah, with a public repository, reproducible pipeline documentation, and a pre-registered evaluation protocol. It is built on open models and open data. The codebase is modular: the embedding layer, fingerprint layer, and classifier are separate components, so a researcher can swap in a different protein language model or a different fingerprint scheme without rewriting the pipeline. That is interoperability by design, not by accident. The Lets Get Started track asks for an early-stage project with a minimal viable product and a clear plan for next steps. TOPOLOGIX has a working prototype, a validated benchmark result, and a defined roadmap. The next steps are three: expand the training set beyond Platinum and SKEMPI 2.0 to include resistance mutations from African pathogen genomics surveillance data, add uncertainty quantification to the classifier output so clinicians can see when a prediction is not trustworthy, and package the tool as a simple command-line interface with a web front end for non-specialist users. The budget request is modest and itemized below. I am a licensed pharmacist with a B.Pharm from the University of Ibadan, an enrolled M.Sc. student in Digital Health at the Hasso Plattner Institute in Potsdam, and an independent researcher with a track record of delivering open, reproducible computational work. I am asking NGI to fund a tool that already works, so that it can reach the people who need it. RESEARCH STATEMENT TOPOLOGIX is an open-source computational tool for predicting drug-resistance mutations from protein sequence alone. The problem it solves is structural: most resistance prediction methods require a resolved three-dimensional protein structure, and most clinically relevant resistance mutations occur in proteins for which no such structure exists. The Platinum benchmark, a standard dataset of 553 resistance mutations, is only about 18 percent covered by structure-based tools. TOPOLOGIX covers 100 percent of mutations in that benchmark because it operates on sequence embeddings, not coordinates. The method has three layers. First, a protein language model, ESM-2, generates delta-embeddings that capture the change in the protein's learned representation when a mutation is introduced. Second, the drug is represented by Morgan/ECFP circular fingerprints, which encode local chemical environments. Third, a Random Forest classifier combines the two representations to predict whether a given mutation confers resistance to a given drug. The pipeline is implemented in Python, uses scikit-learn for the classifier, and is fully reproducible from a public GitHub repository. The results are as follows. On the Platinum benchmark, TOPOLOGIX achieves AUROC 0.804 with a standard deviation of 0.025 across cross-validation folds. On SKEMPI 2.0, a binding-affinity dataset used as a transfer test, it achieves AUROC 0.634. The structure-based baseline mCSM-lig scores approximately 0.70 on comparable tasks but is limited to the subset of mutations with available structures. TOPOLOGIX outperforms that baseline while covering the full mutation set. This is a change in what the tool can do: predict resistance for any sequenced mutation, not just the minority with structural data. The project's history is part of its rigor. Before building TOPOLOGIX, I tested whether interface topology, measured with bipartite persistent homology using an opposition-distance metric implemented in Ripser and GUDHI, could predict drug resistance. The pre-registered, powered replication on the Platinum benchmark produced AUROC 0.425 and 0.485, below chance. I published that negative result and pivoted to a sequence-representation approach. That pivot is documented in the repository. The negative result is not hidden; it is the justification for the current design. The roadmap has three phases. Phase one, dataset expansion: integrate resistance mutation data from African pathogen genomics surveillance, specifically from the GHRU-GSAR work I contributed to as a bioinformatics researcher, to improve generalizability beyond the Western-centric Platinum and SKEMPI benchmarks. Phase two, uncertainty quantification: add conformal prediction or a comparable method to the classifier output so that each prediction carries a calibrated confidence interval. A clinician needs to know when a prediction is unreliable. Phase three, accessibility: package TOPOLOGIX as a command-line tool with a minimal web interface, so that a researcher with no programming background can run a prediction on a sequenced mutation within minutes. The budget request is 12,500 euros. The breakdown: 5,000 euros for compute time for ESM-2 embedding generation and cross-validation on the expanded dataset, 4,000 euros for a part-time research assistant for six months to handle dataset curation and pipeline testing, 2,000 euros for the web interface and packaging work, and 1,500 euros for publication fees and open-access costs for the methods paper. No funds are requested for equipment, travel, or institutional overhead. The project is hosted on GitHub, uses open data, and will remain MIT-licensed. ESSAY: OPEN SOURCE AND COMMUNITY IMPACT The NGI Commons Fund prioritizes software that is open, interoperable, and trustworthy. TOPOLOGIX meets each criterion with a concrete implementation. Openness: the repository is public on GitHub, the license is MIT, and the evaluation protocol was pre-registered before the benchmark was run. Interoperability: the embedding, fingerprint, and classifier layers are separate Python modules with defined interfaces. A user can replace ESM-2 with a newer protein language model, or Morgan fingerprints with a graph-based molecular representation, without rewriting the pipeline. Trustworthiness: the negative result from the earlier topology-based approach is documented in the repository alongside the positive results, so any user can see what was tried, what failed, and why the current design was chosen. The community impact argument is specific to global health equity. Drug-resistance prediction is currently a tool for institutions with structural biology capacity. A hospital or research lab in Nigeria, Kenya, or South Africa that sequences a pathogen genome cannot run mCSM-lig on the resulting mutations because the structures are not available. TOPOLOGIX removes that barrier. The tool runs on a standard laptop, requires no proprietary software, and produces a prediction from sequence data alone. The expansion plan includes integrating resistance data from African pathogen surveillance programs, which means the tool will be trained on the mutations that actually matter in the regions where resistance is most dangerous. This is a specific mechanism by which an open-source tool changes who can participate in resistance-guided drug design. ESSAY: TECHNICAL FEASIBILITY AND DELIVERY CAPACITY The technical risk in TOPOLOGIX is not whether the method works; it is whether the tool can be maintained and extended by a community beyond its original author. The benchmark results are already established. The open question is sustainability. My track record addresses that question directly. I have built and maintained four independent DuckDB-based data pipelines across life sciences, technology, and social science domains, each with documented ingest-to-analyze workflows. I have deployed production systems on Linux VPS infrastructure with systemd, Caddy TLS, CI/CD, and automated backup and disaster-recovery procedures. I have self-hosted local LLM serving with llama.cpp and on-demand model swapping. These are engineering skills. They are the skills required to keep an open-source tool alive after the initial grant period ends. The neurocascade project, a receptor-to-behavior brain-circuit simulation engine, demonstrates my capacity to deliver complex, multi-layer software. It couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics to behavioral readout, with Bayesian calibration via PyMC, and has 62 passing tests out of 62. The ergofluids project, which applies Koopman operator methods with a Mori-Zwanzig memory kernel to drug transport in tumor tissue, demonstrates my commitment to pre-registered validation gates. Its first real-data gate did not meet its primary criterion, and I reported that result directly rather than reframing it. That is the standard of honesty I apply to my own work. TOPOLOGIX will be held to the same standard: pre-registered protocols, public results, and no selective reporting. CHECKLIST - [ ] Confirm NGI Zero Commons Fund eligibility for non-EU applicants; the profile notes a possible EU/associated-country legal entity requirement, verify whether the Hasso Plattner Institute enrollment in Potsdam satisfies this or whether a fiscal sponsor is needed - [ ] Create or confirm the public GitHub repository for TOPOLOGIX is current, with README, MIT license file, and reproducible pipeline documentation - [ ] Upload the pre-registered evaluation protocol for the Platinum benchmark to OSF or Zenodo and link it from the repository - [ ] Prepare a one-page budget justification document with the 12,500 euro breakdown and no overhead line - [ ] Draft a two-sentence project summary for the application form, mentioning AUROC 0.804 on Platinum and 100 percent mutation coverage - [ ] Verify the SKEMPI 2.0 AUROC of 0.634 and the mCSM-lig baseline of approximately 0.70 with the original papers before submission - [ ] Confirm the exact name and spelling of the NGI Zero Commons Fund programme and the Lets Get Started track on the application portal - [ ] Prepare a short bio paragraph mentioning the B.Pharm from University of Ibadan, the M.Sc. enrollment at HPI Potsdam, and the ORCID identifier - [ ] Submit the application through the ossfunded.com portal or the official NGI Zero application system, whichever is specified on the programme page EDITOR NOTES - Eligibility risk: the NGI Zero Commons Fund may require the applicant or a legal entity to be based in the EU or a Horizon Europe associated country. The HPI Potsdam enrollment is a possible anchor, but this must be verified before submission. If it fails, identify a fiscal sponsor or collaborating EU-based institution. - The budget of 12,500 euros is an estimate. The programme page lists amounts up to 50,000 euros, but the Lets Get Started track may have a lower ceiling. Adjust the budget to the track's stated maximum if one is published. - The SKEMPI 2.0 AUROC of 0.634 and the mCSM-lig baseline of approximately 0.70 are taken from the applicant profile. Verify these numbers against the original papers or the project's own benchmark logs before they appear in a final submission. - The essay responses assume the application form asks for short essays on open source impact and technical feasibility. If the form uses different questions, rewrite the essays to match the actual prompts. The motivation letter and research statement are written to the fallback standard and may need trimming if the form imposes word limits.