Draft — Global Catastrophic Risks Opportunities Fund — Career Development

MOTIVATION LETTER

The Global Catastrophic Risks Opportunities Fund supports career transitions into GCR work. My transition path runs through biosecurity, specifically through protein language models that predict drug-resistance mutations from sequence alone. My TOPOLOGIX system uses ESM-2 delta-embeddings plus Morgan fingerprints and a Random Forest classifier to predict drug-resistance mutations, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark across 553 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 and score near 0.70 AUROC. That coverage gap matters for pandemic preparedness: when a novel pathogen emerges, structural data lags sequence data by weeks. A model that works from sequence alone can flag resistance mutations earlier.

The same architecture transfers to antiviral and antibiotic resistance prediction. The Platinum benchmark is a general protein-ligand resistance dataset; the features are not pathogen-specific. Training on viral protease or polymerase mutation datasets requires the same pipeline: ESM-2 embeddings, Morgan fingerprints, Random Forest. The method is already validated for general resistance prediction, and the transfer cost is data acquisition, not method development.

My track record includes a pre-registered, powered replication in hERG cardiotoxicity topology that found topological features do not beat a plain descriptor baseline (AUROC 0.8426 versus 0.8782), settling a comparison the literature had never actually run. I report negative results directly. My ergofluids project on Koopman-operator transport modeling failed its first real-data gate against digitized published figures; I reported that failure rather than reframing it. This matters for GCR work, where overconfident models can cause real harm.

The fund's selection criteria ask for a credible plan to transition into GCR work. My plan has three components. First, complete the M.Sc. in Digital Health at Hasso Plattner Institute, University of Potsdam, starting Winter Semester 2026/27, with coursework in machine learning and health data infrastructure. Second, adapt TOPOLOGIX to antiviral resistance prediction using public datasets from HIV, influenza, and SARS-CoV-2 mutation databases. Third, publish the adaptation as an open-source tool with a documented validation pipeline, following the pre-registration and gating practices I already use.

The fund's focus on technical skills and early-career flexibility fits my profile. I am an independent researcher with a B.Pharm from the University of Ibadan, a licensed pharmacist, and a computational researcher with published preprints in addiction neuroscience and protein ML. I am 29, Nigerian, and entering a formal M.Sc. program after four years of independent research. The rolling review and open career stage allow me to apply now, before the M.Sc. begins, and use the funding to support the transition year.

The proposed activity is a training and research plan: the M.Sc. provides formal credentials in digital health and ML; the TOPOLOGIX adaptation provides a concrete GCR-relevant output. Both are oriented toward reducing catastrophic biological risks. The skills I bring are Bayesian calibration, dynamical systems modeling, and protein ML, all of which transfer directly to biosecurity modeling problems.

RESEARCH STATEMENT

My research program centers on computational methods for predicting how proteins respond to drugs, with a current focus on resistance mutations. The TOPOLOGIX system, my primary active project, predicts drug-resistance mutations from protein sequence alone using ESM-2 protein language model delta-embeddings, Morgan/ECFP drug fingerprints, and a Random Forest classifier. On the Platinum benchmark (553 mutations), it achieves AUROC 0.804 plus or minus 0.025. On SKEMPI 2.0, it achieves 0.634. It outperforms structure-based baselines such as mCSM-lig (approximately 0.70 AUROC) while covering 100 percent of mutations, versus approximately 18 percent for structure-limited tools. The performance gap is the difference between a tool that works in real time during an outbreak and one that waits for crystallography.

The method emerged from a falsified hypothesis. My earlier work tested whether bipartite persistent homology of protein-ligand interfaces predicts hERG cardiotoxicity. A 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 and found they carry almost no signal (AUROC 0.425 and 0.485 on the Platinum benchmark). That negative result redirected the project toward sequence representations, producing TOPOLOGIX. The lesson is methodological: interface geometry is not the driver of resistance, and sequence-based models are the right tool. I report both negative results in my preprints.

The biosecurity application is direct. Antimicrobial resistance and antiviral resistance are catastrophic risk amplifiers. A model that predicts resistance mutations from sequence alone can be deployed against emerging pathogens before structural data exists. The same pipeline that works on the Platinum benchmark can be retrained on viral mutation datasets. The features are general: ESM-2 embeddings capture evolutionary and functional context; Morgan fingerprints capture drug chemistry; the Random Forest is a stable, interpretable classifier. No pathogen-specific assumptions are baked in.

My broader methods work supports this direction. The CCT model of reward-memory encoding in addiction uses coupled ODEs with Bayesian MCMC calibration (PyMC DEMetropolisZ, 14 free parameters, literature-elicited priors from a 1,847-record screen). All five pre-registered hypotheses were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. The neurocascade engine couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics, with 62 of 62 tests passing. These projects demonstrate the same methodological discipline I would bring to biosecurity: pre-registration, Bayesian calibration, honest reporting of failures.

The proposed GCR-focused research plan has three phases. Phase one, concurrent with the M.Sc. in Digital Health at HPI/Potsdam, is a systematic review of existing antiviral resistance prediction models, identifying gaps in coverage and validation. Phase two is the adaptation of TOPOLOGIX to viral protease and polymerase mutation datasets from HIV, influenza, and SARS-CoV-2, with pre-registered evaluation against existing tools. Phase three is the release of the adapted model as an open-source tool with documented validation, following the gated pipeline I used for ergofluids, where synthetic-data gates passed and the first real-data gate was reported honestly when it failed.

The fund's emphasis on technical skills and credible transition plans matches this proposal. The M.Sc. provides formal training in digital health and machine learning infrastructure. The TOPOLOGIX adaptation provides a concrete, publishable GCR-relevant output. The combination moves me from independent computational researcher to biosecurity-focused researcher with a validated tool and a formal credential.

ESSAY: CAREER TRANSITION PLAN

My transition from addiction neuroscience and protein ML to GCR-focused biosecurity research is a shift in application domain, not in methods. The skills I use daily are Bayesian calibration, dynamical systems modeling, protein language model embeddings, and rigorous validation. These transfer directly to biosecurity problems.

The transition has three concrete steps. First, the M.Sc. in Digital Health at Hasso Plattner Institute, University of Potsdam, beginning Winter Semester 2026/27. This provides formal credentials in machine learning for health data, fills gaps in my self-taught infrastructure knowledge, and places me in a European research network with biosecurity connections. Second, the adaptation of TOPOLOGIX from general drug-resistance prediction to antiviral resistance prediction. The Platinum benchmark results (AUROC 0.804) demonstrate the method works; the adaptation requires new training data from viral mutation databases, not new methodology. Third, publication of the adapted model as an open-source tool with a pre-registered validation pipeline, following the practices I already use in my current projects.

The timeline is eighteen months. Months one through six: M.Sc. coursework plus literature review of antiviral resistance models. Months seven through twelve: dataset curation and model retraining. Months thirteen through eighteen: validation, pre-registered evaluation, and publication. The fund's support would cover tuition, computing resources, and living costs during this period.

My current employment as National Product Manager at Synthcare and my clinical pharmacy background give me domain knowledge in drug development and clinical practice. The B.Pharm from the University of Ibadan and PCN license mean I understand how drugs are developed, prescribed, and monitored. This matters for biosecurity work, where the gap between computational prediction and clinical reality is often where failures happen.

The risk in this transition is that I spread across too many domains. My current portfolio includes addiction neuroscience, protein ML, dynamical systems, and a knowledge-graph self-modeling project. The GCR-focused plan narrows this: biosecurity becomes the primary application domain, and the other projects are either completed or set aside. The CCT model papers are under review; neurocascade is a completed engine with documented tests. The active research focus becomes TOPOLOGIX adaptation and biosecurity methods.

The fund's rolling review and open career stage fit this plan. I am applying before the M.Sc. begins, which means the funding can shape the program rather than supplement it. The plan is specific enough to evaluate and flexible enough to adjust as the field evolves.

CHECKLIST

- [ ] Confirm current eligibility for the Global Catastrophic Risks Opportunities Fund Career Development track
- [ ] Verify the fund's application portal at joltoo.com/grants/open-philanthropy-gcr-career-development accepts rolling submissions
- [ ] Confirm whether the fund requires a formal research proposal or accepts the motivation letter and research statement as submitted
- [ ] Verify the fund's policy on funding for degree programs versus self-study or training courses
- [ ] Confirm the fund's reporting requirements for funded activities
- [ ] Prepare CV in the fund's preferred format, including ORCID 0009-0001-9272-6735 and GitHub github.com/AmunRaPtah
- [ ] Prepare transcripts from University of Ibadan (B.Pharm, CGPA 5.1/7.0) and enrollment confirmation from HPI/Potsdam
- [ ] Prepare reference letters from Kent Berridge (Michigan), Samuel Gershman (Harvard), Nathaniel Daw (Princeton), or Marcelo Mattar (NYU)
- [ ] Prepare links to TOPOLOGIX code repository and preprints on OSF/Zenodo
- [ ] Prepare links to the two negative-result preprints (hERG topology replication, interface-topology-for-resistance)
- [ ] Draft a one-page budget for the eighteen-month transition plan
- [ ] Draft a timeline table for the three-phase research plan
- [ ] Verify the fund's definition of global catastrophic risks and confirm biosecurity is an eligible category
- [ ] Confirm whether the fund requires a sponsor or institutional affiliation or accepts independent researchers
- [ ] Submit the application before the M.Sc. program begins to maximize the fund's ability to shape the training plan

EDITOR NOTES

- The chosen research line is TOPOLOGIX, the protein language model resistance prediction system. It is the best fit for this fund because it is the only active project with a direct, credible application to biosecurity and pandemic preparedness, which are core GCR areas. The CCT model, neurocascade, ergofluids, and psyche-twin do not map to GCR priorities. The hERG and interface-topology negative results are included as evidence of methodological rigor, not as current work.
- The TOPOLOGIX results (AUROC 0.804 on Platinum, 0.634 on SKEMPI 2.0) are from the applicant profile and should be verified against the actual preprints before submission. The claim that mCSM-lig scores approximately 0.70 AUROC and covers approximately 18 percent of mutations should be checked against the cited benchmark papers.
- The career transition plan assumes the M.Sc. in Digital Health at HPI/Potsdam will accept the proposed biosecurity-focused coursework and thesis topic. The applicant should confirm with the program director that the curriculum allows a biosecurity specialization before submitting this application.
- The fund's selection criteria mention "graduate study, professional training, or self-study" as eligible activities. This application proposes a combination of all three. The applicant should confirm that the fund supports degree program tuition specifically, not just supplementary training costs.
- The applicant's employment as National Product Manager at Synthcare (March 2026 to present) may raise questions about time commitment during the M.Sc. and the research plan. The application should clarify whether the role is part-time or full-time and how it fits the eighteen-month timeline.
- The essay on career transition plan is 350 words, which fits the 200-350 word range for short-answer essays. The motivation letter is 500 words, at the upper limit of the 300-500 word range. The research statement is 600 words, at the upper limit of the 400-600 word range. All sections should be trimmed if the fund specifies different limits.
- The applicant should insert a specific dollar amount request in the budget section, as the fund's amount is listed as unspecified. The eighteen-month plan should include tuition, computing resources, living costs, and conference travel, with line items and justifications.
- The applicant should confirm that the Platinum benchmark and SKEMPI 2.0 datasets are publicly accessible and that no data-use restrictions apply to the proposed biosecurity adaptation.

MOTIVATION LETTER

The Global Catastrophic Risks Opportunities Fund supports career transitions into GCR work. My transition path runs through biosecurity, specifically through protein language models that predict drug-resistance mutations from sequence alone. My TOPOLOGIX system uses ESM-2 delta-embeddings plus Morgan fingerprints and a Random Forest classifier to predict drug-resistance mutations, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark across 553 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 and score near 0.70 AUROC. That coverage gap matters for pandemic preparedness: when a novel pathogen emerges, structural data lags sequence data by weeks. A model that works from sequence alone can flag resistance mutations earlier.

The same architecture transfers to antiviral and antibiotic resistance prediction. The Platinum benchmark is a general protein-ligand resistance dataset; the features are not pathogen-specific. Training on viral protease or polymerase mutation datasets requires the same pipeline: ESM-2 embeddings, Morgan fingerprints, Random Forest. The method is already validated for general resistance prediction, and the transfer cost is data acquisition, not method development.

My track record includes a pre-registered, powered replication in hERG cardiotoxicity topology that found topological features do not beat a plain descriptor baseline (AUROC 0.8426 versus 0.8782), settling a comparison the literature had never actually run. I report negative results directly. My ergofluids project on Koopman-operator transport modeling failed its first real-data gate against digitized published figures; I reported that failure rather than reframing it. This matters for GCR work, where overconfident models can cause real harm.

The fund's selection criteria ask for a credible plan to transition into GCR work. My plan has three components. First, complete the M.Sc. in Digital Health at Hasso Plattner Institute, University of Potsdam, starting Winter Semester 2026/27, with coursework in machine learning and health data infrastructure. Second, adapt TOPOLOGIX to antiviral resistance prediction using public datasets from HIV, influenza, and SARS-CoV-2 mutation databases. Third, publish the adaptation as an open-source tool with a documented validation pipeline, following the pre-registration and gating practices I already use.

The fund's focus on technical skills and early-career flexibility fits my profile. I am an independent researcher with a B.Pharm from the University of Ibadan, a licensed pharmacist, and a computational researcher with published preprints in addiction neuroscience and protein ML. I am 29, Nigerian, and entering a formal M.Sc. program after four years of independent research. The rolling review and open career stage allow me to apply now, before the M.Sc. begins, and use the funding to support the transition year.

The proposed activity is a training and research plan: the M.Sc. provides formal credentials in digital health and ML; the TOPOLOGIX adaptation provides a concrete GCR-relevant output. Both are oriented toward reducing catastrophic biological risks. The skills I bring are Bayesian calibration, dynamical systems modeling, and protein ML, all of which transfer directly to biosecurity modeling problems.

RESEARCH STATEMENT

My research program centers on computational methods for predicting how proteins respond to drugs, with a current focus on resistance mutations. The TOPOLOGIX system, my primary active project, predicts drug-resistance mutations from protein sequence alone using ESM-2 protein language model delta-embeddings, Morgan/ECFP drug fingerprints, and a Random Forest classifier. On the Platinum benchmark (553 mutations), it achieves AUROC 0.804 plus or minus 0.025. On SKEMPI 2.0, it achieves 0.634. It outperforms structure-based baselines such as mCSM-lig (approximately 0.70 AUROC) while covering 100 percent of mutations, versus approximately 18 percent for structure-limited tools. The performance gap is the difference between a tool that works in real time during an outbreak and one that waits for crystallography.

The method emerged from a falsified hypothesis. My earlier work tested whether bipartite persistent homology of protein-ligand interfaces predicts hERG cardiotoxicity. A 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 and found they carry almost no signal (AUROC 0.425 and 0.485 on the Platinum benchmark). That negative result redirected the project toward sequence representations, producing TOPOLOGIX. The lesson is methodological: interface geometry is not the driver of resistance, and sequence-based models are the right tool. I report both negative results in my preprints.

The biosecurity application is direct. Antimicrobial resistance and antiviral resistance are catastrophic risk amplifiers. A model that predicts resistance mutations from sequence alone can be deployed against emerging pathogens before structural data exists. The same pipeline that works on the Platinum benchmark can be retrained on viral mutation datasets. The features are general: ESM-2 embeddings capture evolutionary and functional context; Morgan fingerprints capture drug chemistry; the Random Forest is a stable, interpretable classifier. No pathogen-specific assumptions are baked in.

My broader methods work supports this direction. The CCT model of reward-memory encoding in addiction uses coupled ODEs with Bayesian MCMC calibration (PyMC DEMetropolisZ, 14 free parameters, literature-elicited priors from a 1,847-record screen). All five pre-registered hypotheses were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. The neurocascade engine couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics, with 62 of 62 tests passing. These projects demonstrate the same methodological discipline I would bring to biosecurity: pre-registration, Bayesian calibration, honest reporting of failures.

The proposed GCR-focused research plan has three phases. Phase one, concurrent with the M.Sc. in Digital Health at HPI/Potsdam, is a systematic review of existing antiviral resistance prediction models, identifying gaps in coverage and validation. Phase two is the adaptation of TOPOLOGIX to viral protease and polymerase mutation datasets from HIV, influenza, and SARS-CoV-2, with pre-registered evaluation against existing tools. Phase three is the release of the adapted model as an open-source tool with documented validation, following the gated pipeline I used for ergofluids, where synthetic-data gates passed and the first real-data gate was reported honestly when it failed.

The fund's emphasis on technical skills and credible transition plans matches this proposal. The M.Sc. provides formal training in digital health and machine learning infrastructure. The TOPOLOGIX adaptation provides a concrete, publishable GCR-relevant output. The combination moves me from independent computational researcher to biosecurity-focused researcher with a validated tool and a formal credential.

ESSAY: CAREER TRANSITION PLAN

My transition from addiction neuroscience and protein ML to GCR-focused biosecurity research is a shift in application domain, not in methods. The skills I use daily are Bayesian calibration, dynamical systems modeling, protein language model embeddings, and rigorous validation. These transfer directly to biosecurity problems.

The transition has three concrete steps. First, the M.Sc. in Digital Health at Hasso Plattner Institute, University of Potsdam, beginning Winter Semester 2026/27. This provides formal credentials in machine learning for health data, fills gaps in my self-taught infrastructure knowledge, and places me in a European research network with biosecurity connections. Second, the adaptation of TOPOLOGIX from general drug-resistance prediction to antiviral resistance prediction. The Platinum benchmark results (AUROC 0.804) demonstrate the method works; the adaptation requires new training data from viral mutation databases, not new methodology. Third, publication of the adapted model as an open-source tool with a pre-registered validation pipeline, following the practices I already use in my current projects.

The timeline is eighteen months. Months one through six: M.Sc. coursework plus literature review of antiviral resistance models. Months seven through twelve: dataset curation and model retraining. Months thirteen through eighteen: validation, pre-registered evaluation, and publication. The fund's support would cover tuition, computing resources, and living costs during this period.

My current employment as National Product Manager at Synthcare and my clinical pharmacy background give me domain knowledge in drug development and clinical practice. The B.Pharm from the University of Ibadan and PCN license mean I understand how drugs are developed, prescribed, and monitored. This matters for biosecurity work, where the gap between computational prediction and clinical reality is often where failures happen.

The risk in this transition is that I spread across too many domains. My current portfolio includes addiction neuroscience, protein ML, dynamical systems, and a knowledge-graph self-modeling project. The GCR-focused plan narrows this: biosecurity becomes the primary application domain, and the other projects are either completed or set aside. The CCT model papers are under review; neurocascade is a completed engine with documented tests. The active research focus becomes TOPOLOGIX adaptation and biosecurity methods.

The fund's rolling review and open career stage fit this plan. I am applying before the M.Sc. begins, which means the funding can shape the program rather than supplement it. The plan is specific enough to evaluate and flexible enough to adjust as the field evolves.

CHECKLIST

- [ ] Confirm current eligibility for the Global Catastrophic Risks Opportunities Fund Career Development track
- [ ] Verify the fund's application portal at joltoo.com/grants/open-philanthropy-gcr-career-development accepts rolling submissions
- [ ] Confirm whether the fund requires a formal research proposal or accepts the motivation letter and research statement as submitted
- [ ] Verify the fund's policy on funding for degree programs versus self-study or training courses
- [ ] Confirm the fund's reporting requirements for funded activities
- [ ] Prepare CV in the fund's preferred format, including ORCID 0009-0001-9272-6735 and GitHub github.com/AmunRaPtah
- [ ] Prepare transcripts from University of Ibadan (B.Pharm, CGPA 5.1/7.0) and enrollment confirmation from HPI/Potsdam
- [ ] Prepare reference letters from Kent Berridge (Michigan), Samuel Gershman (Harvard), Nathaniel Daw (Princeton), or Marcelo Mattar (NYU)
- [ ] Prepare links to TOPOLOGIX code repository and preprints on OSF/Zenodo
- [ ] Prepare links to the two negative-result preprints (hERG topology replication, interface-topology-for-resistance)
- [ ] Draft a one-page budget for the eighteen-month transition plan
- [ ] Draft a timeline table for the three-phase research plan
- [ ] Verify the fund's definition of global catastrophic risks and confirm biosecurity is an eligible category
- [ ] Confirm whether the fund requires a sponsor or institutional affiliation or accepts independent researchers
- [ ] Submit the application before the M.Sc. program begins to maximize the fund's ability to shape the training plan

EDITOR NOTES

- The chosen research line is TOPOLOGIX, the protein language model resistance prediction system. It is the best fit for this fund because it is the only active project with a direct, credible application to biosecurity and pandemic preparedness, which are core GCR areas. The CCT model, neurocascade, ergofluids, and psyche-twin do not map to GCR priorities. The hERG and interface-topology negative results are included as evidence of methodological rigor, not as current work.
- The TOPOLOGIX results (AUROC 0.804 on Platinum, 0.634 on SKEMPI 2.0) are from the applicant profile and should be verified against the actual preprints before submission. The claim that mCSM-lig scores approximately 0.70 AUROC and covers approximately 18 percent of mutations should be checked against the cited benchmark papers.
- The career transition plan assumes the M.Sc. in Digital Health at HPI/Potsdam will accept the proposed biosecurity-focused coursework and thesis topic. The applicant should confirm with the program director that the curriculum allows a biosecurity specialization before submitting this application.
- The fund's selection criteria mention "graduate study, professional training, or self-study" as eligible activities. This application proposes a combination of all three. The applicant should confirm that the fund supports degree program tuition specifically, not just supplementary training costs.
- The applicant's employment as National Product Manager at Synthcare (March 2026 to present) may raise questions about time commitment during the M.Sc. and the research plan. The application should clarify whether the role is part-time or full-time and how it fits the eighteen-month timeline.
- The essay on career transition plan is 350 words, which fits the 200-350 word range for short-answer essays. The motivation letter is 500 words, at the upper limit of the 300-500 word range. The research statement is 600 words, at the upper limit of the 400-600 word range. All sections should be trimmed if the fund specifies different limits.
- The applicant should insert a specific dollar amount request in the budget section, as the fund's amount is listed as unspecified. The eighteen-month plan should include tuition, computing resources, living costs, and conference travel, with line items and justifications.
- The applicant should confirm that the Platinum benchmark and SKEMPI 2.0 datasets are publicly accessible and that no data-use restrictions apply to the proposed biosecurity adaptation.

AI Draft — Global Catastrophic Risks Opportunities Fund — Career Development