MOTIVATION LETTER The Google Conference Scholarship for Africa exists to put African researchers in rooms where the field moves forward. I am applying for support to present my TOPOLOGIX work at a major machine learning or bioinformatics conference, most likely NeurIPS or ISMB, where the intersection of protein language models and drug resistance is an active discussion. My research directly addresses a problem that structure-based tools have failed to solve: predicting drug-resistance mutations from sequence alone. TOPOLOGIX uses ESM-2 protein-language-model delta-embeddings combined with Morgan/ECFP drug fingerprints and a Random Forest classifier. On the Platinum benchmark of 553 mutations, it achieves an AUROC of 0.804 plus or minus 0.025, and 0.634 on SKEMPI 2.0. This beats the structure-based baseline mCSM-lig at approximately 0.70 while covering 100 percent of mutations, compared to roughly 18 percent for structure-limited tools. The method works where structural data is unavailable, which is the majority of real-world clinical cases. The path to TOPOLOGIX included a negative result that shaped its design. My earlier pre-registered study on bipartite persistent homology for hERG cardiotoxicity found that topological features did not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. A follow-up applying the same topological constructs to drug-resistance prediction found almost no signal, AUROC 0.425 and 0.485 on the Platinum benchmark. Those falsifications ruled out interface geometry as the driver and motivated the sequence-representation approach that became TOPOLOGIX. I report these results directly because they are the evidence that the current method is built on tested ground. I am enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and the University of Potsdam, starting Winter Semester 2026/27. I hold a B.Pharm from the University of Ibadan with a German equivalent grade of 1.9 and am a PCN-licensed pharmacist. My research is independent, self-funded, and conducted from Nigeria. I have no institutional travel budget and no alternative funding source for international conference attendance. Attending the conference would serve three concrete purposes. First, presenting TOPOLOGIX to researchers working on protein language models and antimicrobial resistance would put the method in front of the people best positioned to extend it or identify its weaknesses. Second, the conference would allow direct conversations with potential collaborators on scaling the approach to larger benchmarks and clinical datasets. Third, as an independent researcher without a lab or departmental network, these conferences are the primary mechanism by which I can enter the collaboration graph of my field. The scholarship criteria ask for academic merit, relevance, impact, and financial need. My pre-registered, falsification-driven methodology and the TOPOLOGIX results speak to merit. The alignment with AI/ML and digital health speaks to relevance. The gap between sequence-based and structure-based coverage speaks to impact. My independent, self-funded status in Nigeria speaks to need. RESEARCH STATEMENT My research program centers on computational methods for drug response prediction, with a current focus on sequence-based prediction of drug-resistance mutations. The work spans three phases: a falsified topological hypothesis, a validated sequence-based alternative, and an ongoing effort to scale the method to clinical relevance. The first phase tested whether bipartite persistent homology, using an opposition-distance metric implemented with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. This was a pre-registered, powered replication. The result was negative: topological features achieved an AUROC of 0.8426 against 0.8782 for a plain descriptor baseline. The published literature had never actually run this comparison. I ran it and reported the outcome. The second phase applied the same topological constructs to drug-resistance prediction on the Platinum benchmark. The result was again negative: AUROC of 0.425 and 0.485. Interface geometry carried almost no signal for this task. This ruled out the structural hypothesis and motivated a different representation. The third phase, TOPOLOGIX, is the current work. It uses ESM-2 protein-language-model delta-embeddings for the mutation context, Morgan/ECFP fingerprints for the drug, and a Random Forest classifier. On the Platinum benchmark of 553 mutations, TOPOLOGIX achieves an AUROC of 0.804 plus or minus 0.025. On SKEMPI 2.0, it achieves 0.634. The structure-based baseline mCSM-lig scores approximately 0.70 but covers only 18 percent of mutations because it requires structural data. TOPOLOGIX covers 100 percent. For clinical applications where structures are rarely available, coverage is the binding constraint. The methodological stance across all three phases is consistent: pre-register the hypothesis, specify the primary criterion in advance, and report the outcome regardless of direction. My ergofluids work on Koopman-operator methods for drug-vehicle transport follows the same discipline. The synthetic-data gates passed; the first real-data gate against digitized published figures did not meet its pre-registered primary criterion. I reported that directly rather than reframing it. This is the standard I hold myself to, and it is the standard I would bring to any collaboration. The immediate next step for TOPOLOGIX is validation on larger and more diverse benchmarks, followed by prospective testing on clinically observed resistance mutations. The method's dependence on sequence alone makes it applicable to pathogens where structural data does not exist. Scaling the approach requires computational resources and collaborators, both of which are scarce for an independent researcher based in Nigeria. My training supports this work. I hold a B.Pharm from the University of Ibadan, CGPA 5.1/7.0, German equivalent 1.9, and am enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and the University of Potsdam. My computational toolkit includes Python with scipy, numpy, PyMC for Bayesian calibration, RDKit for cheminformatics, and experience with HPC workflows via Nextflow and SLURM. I have built four independent DuckDB-based ingest-to-analyze pipelines across life-sciences and other domains, and I self-host local LLM serving for research use. The conference I am seeking funding to attend is the venue where this work should be presented. The combination of protein language models, drug fingerprints, and a falsification-driven methodology is a contribution to the machine learning for biology community. Presenting there would put TOPOLOGIX in front of the researchers who can stress-test it, extend it, or apply it to their own resistance problems. EDITOR NOTES - Research line selected: TOPOLOGIX, the sequence-based drug-resistance prediction work. This is the only active line that matches Google's AI/ML and digital health focus while having concrete, current results. The hERG and resistance topology studies are included as falsified prior work that motivated TOPOLOGIX, not as ongoing claims. - Eligibility risk: The scholarship targets African-based researchers, and the applicant is Nigeria-based and independent. However, enrollment at HPI/Potsdam may complicate the "based in Africa" criterion. Verify whether the scholarship requires current residence in Africa or only citizenship. If residence is required, confirm that independent research from Nigeria satisfies it despite the upcoming German enrollment. - Conference selection: The letter names NeurIPS and ISMB as target conferences but does not commit to one. The applicant must decide which conference to apply for and confirm that the scholarship covers that specific event. ISMB is the safer fit for bioinformatics; NeurIPS is the stronger fit for the ML methods angle. This choice affects the abstract submission and the supervisor letter. - 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 baseline (approximately 0.70) must be checked against the current preprint versions. The coverage figures (100 percent versus 18 percent) must match the exact benchmark definitions used in the paper. - Missing personal detail: The application requires a supervisor or mentor letter. The applicant has endorsements from Kent Berridge, Samuel Gershman, Nathaniel Daw, and Marcelo Mattar, but no formal supervisor. The applicant must identify which of these contacts, or another collaborator, can write the letter and confirm their willingness before submission. - Financial need documentation: The letter states there is no institutional travel budget and no alternative funding. The applicant should prepare a one-page budget estimate for the conference (registration, flights, accommodation, visa) to attach if the application requests it. - Abstract submission: The scholarship likely requires a submitted or accepted abstract. The applicant must confirm whether the conference deadline for abstract submission precedes the scholarship deadline and plan accordingly. If the abstract is not yet submitted, that is the immediate action item. CHECKLIST - [ ] Confirm scholarship eligibility: African citizenship versus African residence, and whether HPI/Potsdam enrollment affects the "based in Africa" criterion - [ ] Decide target conference: NeurIPS or ISMB, and verify the scholarship covers that event - [ ] Submit abstract to the chosen conference for the TOPOLOGIX work - [ ] Verify TOPOLOGIX AUROC values and coverage figures against current preprint versions - [ ] Identify and confirm a mentor or supervisor willing to write a reference letter - [ ] Prepare a one-page conference travel budget (registration, flights, accommodation, visa) - [ ] Update CV to include TOPOLOGIX results and M.Sc. Digital Health enrollment - [ ] Complete the online application form at the scholarship provider's website - [ ] Submit the motivation letter and research statement as prepared - [ ] Confirm the application deadline from the programme website and submit before it