Draft — Coefficient Giving - Global Catastrophic Risks Opportunities Fund

MOTIVATION LETTER

The Coefficient Giving Global Catastrophic Risks Opportunities Fund exists to move researchers into work that reduces existential risk. My application rests on one specific claim: the methods I have built for predicting drug-resistance mutations from protein sequence alone can be redirected toward biosecurity, specifically the anticipation of antimicrobial resistance and engineered pathogen threats. The core tool, TOPOLOGIX, already achieves an AUROC of 0.804 plus or minus 0.025 on the Platinum benchmark of 553 resistance mutations, and 0.634 on SKEMPI 2.0. It covers 100 percent of mutations tested, where structure-based tools like mCSM-lig cover roughly 18 percent because they require crystallographic structures that do not exist for most emerging threats.

The gap this fund targets is real. Standard biosecurity tools depend on structural data that is unavailable for novel or engineered pathogens. Sequence data, by contrast, is cheap, fast, and universal. My work demonstrates that a protein language model, ESM-2 delta-embeddings, combined with Morgan fingerprints and a Random Forest classifier, can predict resistance from sequence alone. That capability is directly applicable to pandemic preparedness: identifying whether a circulating or engineered strain will evade existing therapeutics before it spreads.

My path into this field is unconventional but relevant. I am a licensed pharmacist with a B.Pharm from the University of Ibadan, a computational researcher with four years of independent work in protein ML and dynamical systems, and an incoming M.Sc. student in Digital Health at the Hasso Plattner Institute and University of Potsdam for Winter Semester 2026/27. The funding would support that degree and the transition it enables: from independent research to a structured position in biosecurity-focused computational biology.

I have already demonstrated the rigor this field requires. My cardiotoxicity topology study, pre-registered and powered, found that persistent homology features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I reported that negative result directly. My ergofluids project, a Koopman-operator method for drug transport, failed its first real-data validation gate; I reported that too, rather than reframing it. Biosecurity decisions cannot afford researchers who hide negative results. I have a track record of not doing so.

The neglectedness criterion is met. Antimicrobial resistance is projected to kill ten million people per year by 2050, yet the computational tools for predicting resistance mutations remain structurally limited. My sequence-based approach addresses a bottleneck that structural biology cannot. The tractability criterion is met by my existing benchmark results. The career-stage flexibility criterion is met by my enrollment in a relevant M.Sc. program and my demonstrated capacity for independent, publication-grade research.

I am applying for this fund because it explicitly supports career transitions into GCR-relevant work. That is precisely what this grant would enable: a pharmacist with a working protein-ML pipeline, a record of honest reporting, and a concrete plan to apply both to biosecurity.

RESEARCH STATEMENT

My research program centers on one question: can we predict how pathogens evolve resistance to drugs from sequence data alone, fast enough and accurately enough to inform pandemic response? The answer, based on my current results, is yes, with caveats that define the next phase of work.

The TOPOLOGIX system, which I designed and built as an independent researcher, combines ESM-2 protein language model delta-embeddings with Morgan and ECFP drug fingerprints, fed into a Random Forest classifier. On the Platinum benchmark of 553 resistance mutations, it achieves an AUROC of 0.804 with a standard deviation of 0.025. On SKEMPI 2.0, it achieves 0.634. These numbers matter because they beat structure-based baselines, mCSM-lig sits near 0.70, while covering the full mutation space. Structure-based tools require crystallographic data that exists for only a fraction of clinically relevant targets. Sequence-based methods do not.

The path to this result included a deliberate negative finding. My earlier work tested whether bipartite persistent homology, an opposition-distance metric computed with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication found it could not: AUROC 0.8426 versus 0.8782 for a plain descriptor baseline. 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. That negative result motivated the pivot to sequence representations, which produced TOPOLOGIX. The method I am proposing to apply to biosecurity exists because I tested and rejected a more elegant alternative first.

The next phase has three components. First, expand TOPOLOGIX beyond the Platinum and SKEMPI benchmarks to viral targets, starting with influenza neuraminidase and SARS-CoV-2 spike protein resistance datasets. Second, integrate the classifier with a pharmacokinetic layer, drawing on my neurocascade work which couples receptor binding to circuit dynamics, to predict not just whether a mutation confers resistance but whether it does so at clinically relevant drug concentrations. Third, build an open pipeline that accepts raw sequence data and returns resistance predictions within hours, deployable in low-resource settings where sequencing capacity is growing but structural biology is not.

My training supports this program. I hold a B.Pharm from the University of Ibadan with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9, and I am a PCN-licensed pharmacist. My computational toolkit includes Python with scipy, numpy, and PyMC for Bayesian calibration, RDKit for cheminformatics, AlphaFold for structure prediction when available, and production infrastructure skills including DuckDB-based pipelines and self-hosted LLM serving. I have built four independent ingest-to-analyze pipelines across life sciences and other domains. I know how to move from raw data to published result.

The biosecurity application is not a stretch. Antimicrobial resistance is a documented catastrophic risk, and the tools for anticipating it are structurally limited by their dependence on structural data. My method removes that dependence. The funding from Coefficient Giving would support my M.Sc. in Digital Health at HPI and Potsdam, where I will formalize my computational training, and would fund the compute and dataset acquisition required to extend TOPOLOGIX to viral targets. The result would be a validated, open tool for predicting resistance in emerging pathogens, built by a researcher who has already demonstrated both the capability and the honesty the field requires.

SHORT-ANSWER ESSAY: CAREER TRANSITION PLAN

My transition from independent computational researcher to biosecurity-focused scientist has three stages, each with a concrete deliverable.

Stage one, the current academic year, is formalization. I am enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and University of Potsdam, starting Winter Semester 2026/27. The degree provides structured training in health data science, complementing my self-taught computational background. The deliverable is a thesis project applying TOPOLOGIX to a viral resistance dataset, supervised within the HPI network.

Stage two, the following year, is domain expansion. I will extend TOPOLOGIX from bacterial resistance benchmarks to viral targets, specifically influenza neuraminidase and SARS-CoV-2 spike protein. The deliverable is a preprint reporting AUROC on these new benchmarks, plus a public code repository with a documented API.

Stage three, post-degree, is placement. I will apply to biosecurity research groups and labs, including those funded by GCR-focused organizations, with a portfolio that includes the viral-resistance results and the open pipeline. The deliverable is a postdoctoral or research scientist position in a biosecurity lab.

The funding from Coefficient Giving is the enabling condition for stage one. It covers tuition and living costs for the M.Sc., freeing me from the clinical pharmacy work I currently do to fund my research. It also covers compute costs for the viral benchmark expansion, which I currently cannot afford at scale. Without this funding, the transition takes longer and may not happen at all; with it, I have a clear two-year path to a biosecurity research position.

SHORT-ANSWER ESSAY: RELEVANT EXPERIENCE AND SKILLS

The relevant experience for this application is not my pharmacology degree, though it helps. It is the sequence of computational projects where I built methods, tested them honestly, and reported results whether they worked or not.

TOPOLOGIX is the direct evidence. It is a working protein-ML pipeline, AUROC 0.804 on Platinum, that predicts drug resistance from sequence alone. It was built after I falsified my own preferred approach: persistent homology for resistance prediction returned AUROC 0.425 and 0.485, and I published that result rather than burying it. That pattern, build, test, report, is the core skill biosecurity needs.

My technical toolkit is broad enough to execute the plan. Python with scipy, numpy, and PyMC for Bayesian calibration; RDKit for molecular fingerprints; ESM-2 for protein embeddings; Ripser and GUDHI for topology when it is the right tool; and production infrastructure skills including DuckDB pipelines, Linux VPS operations, and CI/CD. I have built four independent data pipelines from ingest to analysis. I can deploy a model as a service, which is what an open resistance-prediction tool requires.

My domain knowledge includes pharmacology, which matters for interpreting resistance predictions. I am a licensed pharmacist with clinical experience at Ramset Pharmacy and research experience in AMR genomics at GHRU-GSAR, where I worked on surveillance pipelines. I understand drug mechanisms, dosing, and the clinical context in which resistance emerges. That combination, computational method-building plus pharmacological domain knowledge, is rare and directly applicable to biosecurity.

CHECKLIST

- [ ] Confirm eligibility for Coefficient Giving Global Catastrophic Risks Opportunities Fund as an independent researcher and incoming M.Sc. student
- [ ] Verify application deadline of 2026-08-10 and submit before that date
- [ ] Prepare CV in the format required by the fund, including ORCID 0009-0001-9272-6735 and GitHub github.com/AmunRaPtah
- [ ] Obtain or confirm letters of recommendation from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar, prioritizing Gershman and Daw for GCR relevance
- [ ] Prepare TOPOLOGIX benchmark results as a supplementary document, including Platinum AUROC 0.804 and SKEMPI 2.0 AUROC 0.634
- [ ] Prepare documentation of negative results, including the hERG topology study and the Platinum topology failure, as evidence of reporting rigor
- [ ] Confirm M.Sc. enrollment details at Hasso Plattner Institute and University of Potsdam, including start date and program structure
- [ ] Draft a budget for the funding request, including tuition, living costs, and compute for viral benchmark expansion
- [ ] Identify specific viral resistance datasets for the proposed expansion, including influenza neuraminidase and SARS-CoV-2 spike protein datasets
- [ ] Prepare a one-page project timeline for the two-year transition plan
- [ ] Submit application through the URL https://coefficientgiving.org/apply-for-funding/

EDITOR NOTES

- Eligibility risk: The applicant is Nigerian and currently based in Nigeria, but the M.Sc. is in Germany. Confirm whether the fund has geographic restrictions or preferences, and whether enrollment in a German program affects LMIC-track eligibility.
- Verification needed: The profile lists endorsements from Berridge, Gershman, Daw, and Mattar. Confirm which of these are formal letters versus informal endorsements, and whether any have agreed to write for this specific application.
- Gap to fill: The application does not specify a funding amount. The applicant must insert a concrete budget figure based on German tuition, living costs, and compute estimates, and should verify whether the fund has a stated range.
- Gap to fill: The viral resistance datasets for influenza and SARS-CoV-2 are named but not specified. The applicant should identify specific public datasets with citation details before submission.
- Framing note: The CCT, neurocascade, and psyche-twin research lines are excluded from this application by design. They are not GCR-relevant and including them would dilute the biosecurity angle. Do not add them.

MOTIVATION LETTER

The Coefficient Giving Global Catastrophic Risks Opportunities Fund exists to move researchers into work that reduces existential risk. My application rests on one specific claim: the methods I have built for predicting drug-resistance mutations from protein sequence alone can be redirected toward biosecurity, specifically the anticipation of antimicrobial resistance and engineered pathogen threats. The core tool, TOPOLOGIX, already achieves an AUROC of 0.804 plus or minus 0.025 on the Platinum benchmark of 553 resistance mutations, and 0.634 on SKEMPI 2.0. It covers 100 percent of mutations tested, where structure-based tools like mCSM-lig cover roughly 18 percent because they require crystallographic structures that do not exist for most emerging threats.

The gap this fund targets is real. Standard biosecurity tools depend on structural data that is unavailable for novel or engineered pathogens. Sequence data, by contrast, is cheap, fast, and universal. My work demonstrates that a protein language model, ESM-2 delta-embeddings, combined with Morgan fingerprints and a Random Forest classifier, can predict resistance from sequence alone. That capability is directly applicable to pandemic preparedness: identifying whether a circulating or engineered strain will evade existing therapeutics before it spreads.

My path into this field is unconventional but relevant. I am a licensed pharmacist with a B.Pharm from the University of Ibadan, a computational researcher with four years of independent work in protein ML and dynamical systems, and an incoming M.Sc. student in Digital Health at the Hasso Plattner Institute and University of Potsdam for Winter Semester 2026/27. The funding would support that degree and the transition it enables: from independent research to a structured position in biosecurity-focused computational biology.

I have already demonstrated the rigor this field requires. My cardiotoxicity topology study, pre-registered and powered, found that persistent homology features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I reported that negative result directly. My ergofluids project, a Koopman-operator method for drug transport, failed its first real-data validation gate; I reported that too, rather than reframing it. Biosecurity decisions cannot afford researchers who hide negative results. I have a track record of not doing so.

The neglectedness criterion is met. Antimicrobial resistance is projected to kill ten million people per year by 2050, yet the computational tools for predicting resistance mutations remain structurally limited. My sequence-based approach addresses a bottleneck that structural biology cannot. The tractability criterion is met by my existing benchmark results. The career-stage flexibility criterion is met by my enrollment in a relevant M.Sc. program and my demonstrated capacity for independent, publication-grade research.

I am applying for this fund because it explicitly supports career transitions into GCR-relevant work. That is precisely what this grant would enable: a pharmacist with a working protein-ML pipeline, a record of honest reporting, and a concrete plan to apply both to biosecurity.

RESEARCH STATEMENT

My research program centers on one question: can we predict how pathogens evolve resistance to drugs from sequence data alone, fast enough and accurately enough to inform pandemic response? The answer, based on my current results, is yes, with caveats that define the next phase of work.

The TOPOLOGIX system, which I designed and built as an independent researcher, combines ESM-2 protein language model delta-embeddings with Morgan and ECFP drug fingerprints, fed into a Random Forest classifier. On the Platinum benchmark of 553 resistance mutations, it achieves an AUROC of 0.804 with a standard deviation of 0.025. On SKEMPI 2.0, it achieves 0.634. These numbers matter because they beat structure-based baselines, mCSM-lig sits near 0.70, while covering the full mutation space. Structure-based tools require crystallographic data that exists for only a fraction of clinically relevant targets. Sequence-based methods do not.

The path to this result included a deliberate negative finding. My earlier work tested whether bipartite persistent homology, an opposition-distance metric computed with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication found it could not: AUROC 0.8426 versus 0.8782 for a plain descriptor baseline. 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. That negative result motivated the pivot to sequence representations, which produced TOPOLOGIX. The method I am proposing to apply to biosecurity exists because I tested and rejected a more elegant alternative first.

The next phase has three components. First, expand TOPOLOGIX beyond the Platinum and SKEMPI benchmarks to viral targets, starting with influenza neuraminidase and SARS-CoV-2 spike protein resistance datasets. Second, integrate the classifier with a pharmacokinetic layer, drawing on my neurocascade work which couples receptor binding to circuit dynamics, to predict not just whether a mutation confers resistance but whether it does so at clinically relevant drug concentrations. Third, build an open pipeline that accepts raw sequence data and returns resistance predictions within hours, deployable in low-resource settings where sequencing capacity is growing but structural biology is not.

My training supports this program. I hold a B.Pharm from the University of Ibadan with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9, and I am a PCN-licensed pharmacist. My computational toolkit includes Python with scipy, numpy, and PyMC for Bayesian calibration, RDKit for cheminformatics, AlphaFold for structure prediction when available, and production infrastructure skills including DuckDB-based pipelines and self-hosted LLM serving. I have built four independent ingest-to-analyze pipelines across life sciences and other domains. I know how to move from raw data to published result.

The biosecurity application is not a stretch. Antimicrobial resistance is a documented catastrophic risk, and the tools for anticipating it are structurally limited by their dependence on structural data. My method removes that dependence. The funding from Coefficient Giving would support my M.Sc. in Digital Health at HPI and Potsdam, where I will formalize my computational training, and would fund the compute and dataset acquisition required to extend TOPOLOGIX to viral targets. The result would be a validated, open tool for predicting resistance in emerging pathogens, built by a researcher who has already demonstrated both the capability and the honesty the field requires.

SHORT-ANSWER ESSAY: CAREER TRANSITION PLAN

My transition from independent computational researcher to biosecurity-focused scientist has three stages, each with a concrete deliverable.

Stage one, the current academic year, is formalization. I am enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and University of Potsdam, starting Winter Semester 2026/27. The degree provides structured training in health data science, complementing my self-taught computational background. The deliverable is a thesis project applying TOPOLOGIX to a viral resistance dataset, supervised within the HPI network.

Stage two, the following year, is domain expansion. I will extend TOPOLOGIX from bacterial resistance benchmarks to viral targets, specifically influenza neuraminidase and SARS-CoV-2 spike protein. The deliverable is a preprint reporting AUROC on these new benchmarks, plus a public code repository with a documented API.

Stage three, post-degree, is placement. I will apply to biosecurity research groups and labs, including those funded by GCR-focused organizations, with a portfolio that includes the viral-resistance results and the open pipeline. The deliverable is a postdoctoral or research scientist position in a biosecurity lab.

The funding from Coefficient Giving is the enabling condition for stage one. It covers tuition and living costs for the M.Sc., freeing me from the clinical pharmacy work I currently do to fund my research. It also covers compute costs for the viral benchmark expansion, which I currently cannot afford at scale. Without this funding, the transition takes longer and may not happen at all; with it, I have a clear two-year path to a biosecurity research position.

SHORT-ANSWER ESSAY: RELEVANT EXPERIENCE AND SKILLS

The relevant experience for this application is not my pharmacology degree, though it helps. It is the sequence of computational projects where I built methods, tested them honestly, and reported results whether they worked or not.

TOPOLOGIX is the direct evidence. It is a working protein-ML pipeline, AUROC 0.804 on Platinum, that predicts drug resistance from sequence alone. It was built after I falsified my own preferred approach: persistent homology for resistance prediction returned AUROC 0.425 and 0.485, and I published that result rather than burying it. That pattern, build, test, report, is the core skill biosecurity needs.

My technical toolkit is broad enough to execute the plan. Python with scipy, numpy, and PyMC for Bayesian calibration; RDKit for molecular fingerprints; ESM-2 for protein embeddings; Ripser and GUDHI for topology when it is the right tool; and production infrastructure skills including DuckDB pipelines, Linux VPS operations, and CI/CD. I have built four independent data pipelines from ingest to analysis. I can deploy a model as a service, which is what an open resistance-prediction tool requires.

My domain knowledge includes pharmacology, which matters for interpreting resistance predictions. I am a licensed pharmacist with clinical experience at Ramset Pharmacy and research experience in AMR genomics at GHRU-GSAR, where I worked on surveillance pipelines. I understand drug mechanisms, dosing, and the clinical context in which resistance emerges. That combination, computational method-building plus pharmacological domain knowledge, is rare and directly applicable to biosecurity.

CHECKLIST

- [ ] Confirm eligibility for Coefficient Giving Global Catastrophic Risks Opportunities Fund as an independent researcher and incoming M.Sc. student
- [ ] Verify application deadline of 2026-08-10 and submit before that date
- [ ] Prepare CV in the format required by the fund, including ORCID 0009-0001-9272-6735 and GitHub github.com/AmunRaPtah
- [ ] Obtain or confirm letters of recommendation from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar, prioritizing Gershman and Daw for GCR relevance
- [ ] Prepare TOPOLOGIX benchmark results as a supplementary document, including Platinum AUROC 0.804 and SKEMPI 2.0 AUROC 0.634
- [ ] Prepare documentation of negative results, including the hERG topology study and the Platinum topology failure, as evidence of reporting rigor
- [ ] Confirm M.Sc. enrollment details at Hasso Plattner Institute and University of Potsdam, including start date and program structure
- [ ] Draft a budget for the funding request, including tuition, living costs, and compute for viral benchmark expansion
- [ ] Identify specific viral resistance datasets for the proposed expansion, including influenza neuraminidase and SARS-CoV-2 spike protein datasets
- [ ] Prepare a one-page project timeline for the two-year transition plan
- [ ] Submit application through the URL https://coefficientgiving.org/apply-for-funding/

EDITOR NOTES

- Eligibility risk: The applicant is Nigerian and currently based in Nigeria, but the M.Sc. is in Germany. Confirm whether the fund has geographic restrictions or preferences, and whether enrollment in a German program affects LMIC-track eligibility.
- Verification needed: The profile lists endorsements from Berridge, Gershman, Daw, and Mattar. Confirm which of these are formal letters versus informal endorsements, and whether any have agreed to write for this specific application.
- Gap to fill: The application does not specify a funding amount. The applicant must insert a concrete budget figure based on German tuition, living costs, and compute estimates, and should verify whether the fund has a stated range.
- Gap to fill: The viral resistance datasets for influenza and SARS-CoV-2 are named but not specified. The applicant should identify specific public datasets with citation details before submission.
- Framing note: The CCT, neurocascade, and psyche-twin research lines are excluded from this application by design. They are not GCR-relevant and including them would dilute the biosecurity angle. Do not add them.

AI Draft — Coefficient Giving - Global Catastrophic Risks Opportunities Fund