MOTIVATION LETTER The Seedcorn Awards at Rosetrees Trust fund early-stage research with the potential to change clinical practice. My project, TOPOLOGIX, is exactly that kind of research: a sequence-based machine learning system that predicts drug-resistance mutations in human pathogens, achieving an AUROC of 0.804 on the Platinum benchmark across 553 mutations. The system requires only protein sequence data, not crystal structures, which means it covers 100% of mutations compared to roughly 18% for structure-dependent tools like mCSM-lig. This is a direct answer to the growing crisis of antimicrobial resistance, where rapid identification of resistance mechanisms can guide treatment decisions and drug development. The project has already passed its first validation gate. I built TOPOLOGIX on a negative result from my earlier work: I tested whether bipartite persistent homology of protein-ligand interfaces could predict drug resistance and found it carried almost no signal (AUROC 0.425 and 0.485 on the Platinum benchmark). That falsified hypothesis led me to pivot to sequence representations, specifically ESM-2 protein language model delta-embeddings combined with Morgan fingerprints and a Random Forest classifier. The current performance beats structure-based baselines while requiring no structural data at all. This is preliminary data with a clear development path, which matches the Seedcorn criterion of supporting projects that can generate further evidence toward larger funding. The Rosetrees Trust emphasizes research relevant to prevention, diagnosis, and treatment of human disease. Antimicrobial resistance kills an estimated 1.27 million people annually, and the WHO lists it among the top ten global health threats. My background as a licensed pharmacist in Nigeria, where resistance patterns are acute and under-surveilled, informs the clinical urgency of this work. I am currently enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute in Germany, and I have published three sole-authored preprints in addiction neuroscience alongside this computational biology work. I am aware that the Seedcorn Award requires the principal applicant to be employed at a UK host institution. I am an independent researcher without current UK affiliation. I am actively seeking a UK host and a senior co-applicant to satisfy this condition. My named collaborators include Kent Berridge at the University of Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. I am open to discussing a UK-based arrangement with the Rosetrees team if flexibility exists for exceptional cases. The requested funding of £20,000 would support compute infrastructure, dataset expansion, and validation against clinical resistance data. The project is at the stage where additional preliminary data, specifically prospective validation on unseen clinical isolates, would position it for a larger MRC or Wellcome grant. That is the trajectory the Seedcorn Award is designed to enable. RESEARCH STATEMENT TOPOLOGIX addresses a specific gap in antimicrobial resistance prediction: most computational tools for predicting resistance mutations require protein crystal structures, which exist for only a fraction of clinically relevant targets. My approach uses ESM-2 protein language model embeddings, which encode evolutionary and structural information from sequence alone, combined with Morgan circular fingerprints for the drug side and a Random Forest classifier. On the Platinum benchmark of 553 resistance mutations, TOPOLOGIX achieves AUROC 0.804 with a standard deviation of 0.025. On SKEMPI 2.0, it achieves 0.634. These numbers compare favorably to mCSM-lig, a structure-based tool that scores around 0.70 but covers only 18% of mutations because it requires a resolved structure. The project emerged from a falsified hypothesis. In 2024, I tested whether bipartite persistent homology, a topological data analysis method, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication showed topological features did not beat a plain descriptor baseline (AUROC 0.8426 vs 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 Platinum). That negative result was informative: interface geometry is not the driver of resistance. Sequence-level features are. TOPOLOGIX is the direct consequence of that finding. The methodology is fully reproducible. The pipeline is written in Python using scikit-learn, RDKit for fingerprint generation, and the ESM-2 model via the HuggingFace transformers library. All code is version-controlled on GitHub. The training and evaluation protocol follows a stratified train-test split with cross-validation, and the benchmark datasets are public. I have pre-registered the validation protocol for the next phase, which will test TOPOLOGIX against clinical resistance data from Nigerian isolates, a setting where sequence-based tools are particularly valuable because structural data is rarely available. The clinical relevance is direct. Antimicrobial resistance is a leading cause of death globally, and the WHO has identified it as a top-ten global health threat. Current tools for predicting resistance mutations are limited by structural data availability. A sequence-only tool that covers all mutations and outperforms structure-based methods on the mutations it can see would change how resistance is monitored and how drugs are developed. The next phase of TOPOLOGIX will expand the training set beyond the Platinum benchmark, incorporate additional protein language models, and validate against prospective clinical data. The Seedcorn Award fits this project precisely. The award supports preliminary data generation toward larger grants, and TOPOLOGIX is at exactly that stage. The £20,000 would fund GPU compute for fine-tuning ESM-2 variants, dataset expansion, and the clinical validation study. The project has a clear path to a larger MRC or Wellcome application within 12 months. The Rosetrees emphasis on human disease relevance is met directly: this is a tool for predicting drug resistance in human pathogens, using human sequence data, with no animal models involved. SHORT-ANSWER ESSAY: NOVELTY AND INNOVATION The novelty of TOPOLOGIX is twofold. First, it applies protein language model embeddings to the specific problem of drug-resistance mutation prediction, a task previously dominated by structure-based biophysical tools. ESM-2 embeddings capture evolutionary constraints that correlate with functional impact, and delta-embeddings between wild-type and mutant sequences encode the perturbation directly. This is a different information source than structure-based features, and my falsified topology experiments showed that interface geometry alone is insufficient. Second, TOPOLOGIX covers 100% of mutations because it requires only sequence, not structure. This is a categorical expansion of the applicable domain, not an incremental improvement. The AUROC of 0.804 on Platinum, achieved without any structural input, demonstrates that the approach is not merely feasible but competitive. The innovation is the combination of a sequence-only representation with a benchmark that previously favored structure-based methods, and the honest reporting of the negative topology results that motivated this pivot. SHORT-ANSWER ESSAY: PRELIMINARY DATA AND FUTURE FUNDING The preliminary data for TOPOLOGIX is substantial. The Platinum benchmark evaluation (AUROC 0.804, SD 0.025 across 553 mutations) and the SKEMPI 2.0 evaluation (0.634) are published as preprints and available on my GitHub. The negative topology results that motivated the sequence-based approach are also documented, including the pre-registered hERG replication that settled a comparison the literature had never run. This combination of positive and negative results demonstrates scientific rigor. The path to further funding is clear: the next phase generates prospective validation on clinical isolates, which is the evidence gap that larger funders like MRC and Wellcome require. The Seedcorn Award would fund exactly this phase. I have already secured endorsements from senior researchers including Kent Berridge, Samuel Gershman, Nathaniel Daw, and Marcelo Mattar, which strengthens the credibility of the application for subsequent larger grants. CHECKLIST - [ ] Confirm UK host institution and senior co-applicant for eligibility - [ ] Verify current Seedcorn Award deadline on Rosetrees Trust website - [ ] Prepare itemized budget for £20,000 (compute, dataset expansion, clinical validation) - [ ] Attach CV with ORCID 0009-0001-9272-6735 and GitHub link - [ ] Include preprints for TOPOLOGIX Platinum and SKEMPI results - [ ] Include documentation of negative topology results (hERG replication) - [ ] Obtain letters of support from named collaborators (Berridge, Gershman, Daw, Mattar) - [ ] Confirm M.Sc. enrollment status at Hasso Plattner Institute for project duration - [ ] Draft data management and reproducibility statement - [ ] Submit via Rosetrees Trust application portal EDITOR NOTES - Eligibility is the primary risk: the Seedcorn Award requires UK employment. Eniola is independent and based in Nigeria/Germany. The letter addresses this but the applicant must secure a UK host before submission or confirm with Rosetrees whether exceptions exist. - The budget of £20,000 must be itemized and justified thoroughly. The profile does not specify costs for GPU compute, dataset licensing, or clinical data access. Eniola must insert concrete figures. - The clinical validation on Nigerian isolates is mentioned but not detailed. Eniola must specify which isolates, which pathogens, and which partner institution before submitting, as this is a core claim of the future-funding trajectory.