Draft — Google DeepMind Artificial Intelligence Masters Scholarship for African Students

MOTIVATION LETTER

The Google DeepMind Artificial Intelligence Masters Scholarship for African Students funds a two-year M.Sc. in Digital Health at the Hasso Plattner Institute / University of Potsdam, where I am enrolled for Winter Semester 2026/27. My research applies machine learning to a problem that kills more people in Africa than any other: antimicrobial resistance. My TOPOLOGIX system predicts drug-resistance mutations from protein sequence alone, using 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, TOPOLOGIX achieves an AUROC of 0.804 plus or minus 0.025, and 0.634 on SKEMPI 2.0. It beats structure-based baselines such as mCSM-lig at approximately 0.70 while covering 100 percent of mutations, compared to roughly 18 percent for structure-limited tools. This matters because most resistance mutations occur in proteins with no experimentally determined structure, and Africa's pathogen burden is disproportionately high.

The DeepMind scholarship's stated mission is to support African students who will use AI to address scientific challenges. My work fits that mission directly. I am a licensed pharmacist (B.Pharm, University of Ibadan, CGPA 5.1/7.0, German equivalent 1.9) with three years of computational research across addiction neuroscience, protein ML, and dynamical systems. My M.Sc. at HPI will deepen my AI expertise in areas the programme explicitly targets: machine learning, data engineering, and AI for science. I will use that training to scale TOPOLOGIX from a research prototype into a validated tool for drug development and public-health surveillance in Nigeria and across the continent.

I have already demonstrated the rigor DeepMind expects. My CCT model, a tripartite pharmacological framework for reward-memory encoding prevention in addiction, is a coupled three-axis ODE model calibrated with Bayesian MCMC (PyMC DEMetropolisZ, 14 free parameters, priors elicited from a 1,847-record literature screen). All five pre-registered hypotheses were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. Three sole-authored preprints are under review at peer-reviewed journals. I have also run pre-registered, powered replication studies that produced negative results, including a cardiotoxicity topology study showing bipartite persistent homology does not beat a plain descriptor baseline (AUROC 0.8426 vs 0.8782), and an interface-topology-for-resistance study that ruled out interface geometry as the driver of resistance prediction (AUROC 0.425 and 0.485). I report negative results directly rather than reframing them. That is the standard DeepMind's research culture demands.

The scholarship covers full tuition and fees for two years, which is precisely the support I need. I am an independent researcher with no institutional funding. My employment history, including National Product Manager at Synthcare and Research Assistant at GHRU-GSAR for AMR genomics, has been self-funded. The DeepMind scholarship would remove the financial barrier to completing my M.Sc. and allow me to focus entirely on building AI tools for African health challenges.

RESEARCH STATEMENT

My research program centers on one question: can machine learning predict how pathogens evolve resistance to drugs, using only sequence data, so that clinicians and drug developers can act before resistance spreads? The answer, demonstrated in my TOPOLOGIX system, is yes, with measurable performance and full coverage of sequence-only mutations.

TOPOLOGIX uses ESM-2 protein language model delta-embeddings to represent the change in a protein's sequence when a mutation occurs. These delta-embeddings capture the functional impact of a mutation without requiring a crystal structure. I combine them with Morgan/ECFP drug fingerprints to represent the ligand, then train a Random Forest classifier to predict whether a given mutation confers resistance. 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. It outperforms structure-based baselines such as mCSM-lig at approximately 0.70 while covering 100 percent of mutations, compared to roughly 18 percent for structure-limited tools. This coverage gap is the core advantage: most clinically relevant mutations occur in proteins without solved structures, and structure-based tools simply cannot score them.

The path to TOPOLOGIX was not linear. My earlier work tested whether topological data analysis, specifically bipartite persistent homology with an opposition-distance metric using Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication found that topological features do 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 carry almost no signal (AUROC 0.425 and 0.485 on the Platinum benchmark). These negative results were published as preprints and reported honestly. They ruled out interface geometry as the driver of resistance and motivated the sequence-representation approach that became TOPOLOGIX. This is how I work: pre-register hypotheses, run powered tests, report what the data says, and pivot when the evidence demands it.

My M.Sc. in Digital Health at HPI/Potsdam will extend this work in three specific ways. First, I will deepen my machine learning fundamentals, particularly in deep learning architectures that could replace the Random Forest with a neural model trained end-to-end on sequence and fingerprint inputs. Second, I will learn health-data engineering at scale, which is required to deploy TOPOLOGIX on real clinical and genomic datasets in Nigeria. Third, I will gain formal training in AI ethics and deployment, which matters when a tool predicts drug resistance that could change treatment decisions.

The broader research context includes two additional lines that demonstrate my multi-domain computational rigor. The CCT model is a tripartite pharmacological framework for reward-memory encoding prevention in addiction, using a coupled three-axis ODE model (dopaminergic RPE, NMDAR-dependent LTP, affective contrast) calibrated with Bayesian MCMC. All five pre-registered hypotheses were confirmed. The neurocascade engine couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics to behavioral readouts, with 62 of 62 tests passing. These projects show I can build and validate complex computational models across neuroscience and pharmacology. But TOPOLOGIX is the project most directly aligned with DeepMind's AI for Science mission, because it applies modern AI methods to a concrete biomedical problem with clear public-health impact in Africa.

The next step for TOPOLOGIX is validation on African pathogen data. Antimicrobial resistance rates in Nigeria are among the highest in the world, and genomic surveillance infrastructure is sparse. I plan to collaborate with the GHRU-GSAR network, where I previously worked on AMR genomics and surveillance pipelines, to test TOPOLOGIX on Nigerian clinical isolates. The DeepMind scholarship would fund the two years of M.Sc. training that makes this deployment possible.

ESSAY: LEADERSHIP AND COMMITMENT TO AI FOR SOCIAL GOOD IN AFRICA

My commitment to using AI for social good in Africa is demonstrated by a decade of work in Nigerian health systems, not by statements of intent. As a clinical pharmacist at Ramset Pharmacy and a National Product Manager at Synthcare, I saw directly how antimicrobial resistance complicates treatment decisions in settings where susceptibility testing is often unavailable. A patient with a resistant infection may receive a drug that does not work, and the clinician has no data to guide the choice. TOPOLOGIX addresses this gap by predicting resistance from sequence data, which is increasingly cheap to generate even where laboratory infrastructure is limited.

I have also built the technical infrastructure to deploy AI tools in resource-constrained settings. I have developed four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech/AI, and social science domains. I self-host local LLM serving with llama.cpp and on-demand model swapping, which means I can run AI models without relying on cloud services that are expensive or unavailable in Nigeria. I operate production Linux VPS systems with systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery. These are not academic skills; they are the practical capabilities required to deploy and maintain AI systems in African health institutions.

My leadership is evidenced by independent research execution. I have designed, pre-registered, and completed multiple computational studies as a sole researcher, including the CCT model with 14 free parameters calibrated via Bayesian MCMC, and the TOPOLOGIX system with its sequence-based resistance prediction. I have secured endorsements from Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. These researchers do not endorse casually; they reviewed my work and found it rigorous.

The DeepMind scholarship's emphasis on AI for social good in Africa matches my existing practice. I am not waiting for a degree to start contributing. I have already built the tools, published the preprints, and reported the negative results honestly. The M.Sc. at HPI will give me the formal AI training and the institutional platform to scale this work, and the DeepMind scholarship will make that training financially possible.

CHECKLIST

- [ ] Complete Google DeepMind AI Masters Scholarship application form at https://deepmind.google/education/
- [ ] Upload motivation letter (this document, 300-500 words)
- [ ] Upload research statement (this document, 400-600 words)
- [ ] Upload essay on leadership and AI for social good in Africa (this document, 200-350 words)
- [ ] Obtain and upload academic transcripts from University of Ibadan (B.Pharm, CGPA 5.1/7.0)
- [ ] Obtain and upload proof of enrollment or admission to M.Sc. Digital Health at HPI/Potsdam for Winter Semester 2026/27
- [ ] Prepare CV listing employment history (Synthcare, Ramset Pharmacy, CDDDP, GHRU-GSAR) and research lines
- [ ] Prepare list of preprints with URLs (OSF/Zenodo for CCT, TOPOLOGIX, cardiotoxicity study, interface-topology study)
- [ ] Prepare ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) links
- [ ] Prepare two reference letters, ideally from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar
- [ ] Verify deadline: 2025-04-30
- [ ] Confirm whether IIE administers this scholarship and whether additional eligibility criteria apply

EDITOR NOTES

- Eligibility risk: The profile lists enrollment in M.Sc. Digital Health at HPI/Potsdam for Winter Semester 2026/27, but the scholarship deadline is 2025-04-30. Verify whether the scholarship requires current enrollment or admission for the upcoming academic year, and whether the program start date aligns with the scholarship's funding period. If the M.Sc. start date is after the scholarship's funding window, this is a critical mismatch that must be resolved before submission.

- Verification needed: Confirm that HPI/Potsdam is a partner university for the DeepMind scholarship. The profile states the program is at a partner-eligible university, but this must be verified against the official partner list. If HPI is not a partner, the application is ineligible regardless of merit.

- Gap to fill: The essay on leadership and AI for social good in Africa is 200-350 words, but the draft above is longer. The applicant must trim it to the required limit. Also, the applicant should insert a specific anecdote or example from their clinical pharmacy work in Nigeria that illustrates the antimicrobial resistance problem, as the current draft references it generally but lacks a concrete patient or clinical scenario.

- Fact check: The profile states TOPOLOGIX beats mCSM-lig at approximately 0.70 AUROC, but the exact comparison methodology should be verified. The applicant should confirm whether mCSM-lig was run on the same Platinum benchmark or a different dataset, as this affects the validity of the comparison claim.

- Honest framing: The cardiotoxicity and interface-topology studies are negative results. The draft presents them as evidence of rigor, which is appropriate, but the applicant must ensure they are not described as current, ongoing work. They are completed, published preprints that motivated the pivot to TOPOLOGIX.

MOTIVATION LETTER

The Google DeepMind Artificial Intelligence Masters Scholarship for African Students funds a two-year M.Sc. in Digital Health at the Hasso Plattner Institute / University of Potsdam, where I am enrolled for Winter Semester 2026/27. My research applies machine learning to a problem that kills more people in Africa than any other: antimicrobial resistance. My TOPOLOGIX system predicts drug-resistance mutations from protein sequence alone, using 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, TOPOLOGIX achieves an AUROC of 0.804 plus or minus 0.025, and 0.634 on SKEMPI 2.0. It beats structure-based baselines such as mCSM-lig at approximately 0.70 while covering 100 percent of mutations, compared to roughly 18 percent for structure-limited tools. This matters because most resistance mutations occur in proteins with no experimentally determined structure, and Africa's pathogen burden is disproportionately high.

The DeepMind scholarship's stated mission is to support African students who will use AI to address scientific challenges. My work fits that mission directly. I am a licensed pharmacist (B.Pharm, University of Ibadan, CGPA 5.1/7.0, German equivalent 1.9) with three years of computational research across addiction neuroscience, protein ML, and dynamical systems. My M.Sc. at HPI will deepen my AI expertise in areas the programme explicitly targets: machine learning, data engineering, and AI for science. I will use that training to scale TOPOLOGIX from a research prototype into a validated tool for drug development and public-health surveillance in Nigeria and across the continent.

I have already demonstrated the rigor DeepMind expects. My CCT model, a tripartite pharmacological framework for reward-memory encoding prevention in addiction, is a coupled three-axis ODE model calibrated with Bayesian MCMC (PyMC DEMetropolisZ, 14 free parameters, priors elicited from a 1,847-record literature screen). All five pre-registered hypotheses were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. Three sole-authored preprints are under review at peer-reviewed journals. I have also run pre-registered, powered replication studies that produced negative results, including a cardiotoxicity topology study showing bipartite persistent homology does not beat a plain descriptor baseline (AUROC 0.8426 vs 0.8782), and an interface-topology-for-resistance study that ruled out interface geometry as the driver of resistance prediction (AUROC 0.425 and 0.485). I report negative results directly rather than reframing them. That is the standard DeepMind's research culture demands.

The scholarship covers full tuition and fees for two years, which is precisely the support I need. I am an independent researcher with no institutional funding. My employment history, including National Product Manager at Synthcare and Research Assistant at GHRU-GSAR for AMR genomics, has been self-funded. The DeepMind scholarship would remove the financial barrier to completing my M.Sc. and allow me to focus entirely on building AI tools for African health challenges.

RESEARCH STATEMENT

My research program centers on one question: can machine learning predict how pathogens evolve resistance to drugs, using only sequence data, so that clinicians and drug developers can act before resistance spreads? The answer, demonstrated in my TOPOLOGIX system, is yes, with measurable performance and full coverage of sequence-only mutations.

TOPOLOGIX uses ESM-2 protein language model delta-embeddings to represent the change in a protein's sequence when a mutation occurs. These delta-embeddings capture the functional impact of a mutation without requiring a crystal structure. I combine them with Morgan/ECFP drug fingerprints to represent the ligand, then train a Random Forest classifier to predict whether a given mutation confers resistance. 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. It outperforms structure-based baselines such as mCSM-lig at approximately 0.70 while covering 100 percent of mutations, compared to roughly 18 percent for structure-limited tools. This coverage gap is the core advantage: most clinically relevant mutations occur in proteins without solved structures, and structure-based tools simply cannot score them.

The path to TOPOLOGIX was not linear. My earlier work tested whether topological data analysis, specifically bipartite persistent homology with an opposition-distance metric using Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. A pre-registered, powered replication found that topological features do 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 carry almost no signal (AUROC 0.425 and 0.485 on the Platinum benchmark). These negative results were published as preprints and reported honestly. They ruled out interface geometry as the driver of resistance and motivated the sequence-representation approach that became TOPOLOGIX. This is how I work: pre-register hypotheses, run powered tests, report what the data says, and pivot when the evidence demands it.

My M.Sc. in Digital Health at HPI/Potsdam will extend this work in three specific ways. First, I will deepen my machine learning fundamentals, particularly in deep learning architectures that could replace the Random Forest with a neural model trained end-to-end on sequence and fingerprint inputs. Second, I will learn health-data engineering at scale, which is required to deploy TOPOLOGIX on real clinical and genomic datasets in Nigeria. Third, I will gain formal training in AI ethics and deployment, which matters when a tool predicts drug resistance that could change treatment decisions.

The broader research context includes two additional lines that demonstrate my multi-domain computational rigor. The CCT model is a tripartite pharmacological framework for reward-memory encoding prevention in addiction, using a coupled three-axis ODE model (dopaminergic RPE, NMDAR-dependent LTP, affective contrast) calibrated with Bayesian MCMC. All five pre-registered hypotheses were confirmed. The neurocascade engine couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics to behavioral readouts, with 62 of 62 tests passing. These projects show I can build and validate complex computational models across neuroscience and pharmacology. But TOPOLOGIX is the project most directly aligned with DeepMind's AI for Science mission, because it applies modern AI methods to a concrete biomedical problem with clear public-health impact in Africa.

The next step for TOPOLOGIX is validation on African pathogen data. Antimicrobial resistance rates in Nigeria are among the highest in the world, and genomic surveillance infrastructure is sparse. I plan to collaborate with the GHRU-GSAR network, where I previously worked on AMR genomics and surveillance pipelines, to test TOPOLOGIX on Nigerian clinical isolates. The DeepMind scholarship would fund the two years of M.Sc. training that makes this deployment possible.

ESSAY: LEADERSHIP AND COMMITMENT TO AI FOR SOCIAL GOOD IN AFRICA

My commitment to using AI for social good in Africa is demonstrated by a decade of work in Nigerian health systems, not by statements of intent. As a clinical pharmacist at Ramset Pharmacy and a National Product Manager at Synthcare, I saw directly how antimicrobial resistance complicates treatment decisions in settings where susceptibility testing is often unavailable. A patient with a resistant infection may receive a drug that does not work, and the clinician has no data to guide the choice. TOPOLOGIX addresses this gap by predicting resistance from sequence data, which is increasingly cheap to generate even where laboratory infrastructure is limited.

I have also built the technical infrastructure to deploy AI tools in resource-constrained settings. I have developed four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech/AI, and social science domains. I self-host local LLM serving with llama.cpp and on-demand model swapping, which means I can run AI models without relying on cloud services that are expensive or unavailable in Nigeria. I operate production Linux VPS systems with systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery. These are not academic skills; they are the practical capabilities required to deploy and maintain AI systems in African health institutions.

My leadership is evidenced by independent research execution. I have designed, pre-registered, and completed multiple computational studies as a sole researcher, including the CCT model with 14 free parameters calibrated via Bayesian MCMC, and the TOPOLOGIX system with its sequence-based resistance prediction. I have secured endorsements from Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. These researchers do not endorse casually; they reviewed my work and found it rigorous.

The DeepMind scholarship's emphasis on AI for social good in Africa matches my existing practice. I am not waiting for a degree to start contributing. I have already built the tools, published the preprints, and reported the negative results honestly. The M.Sc. at HPI will give me the formal AI training and the institutional platform to scale this work, and the DeepMind scholarship will make that training financially possible.

CHECKLIST

- [ ] Complete Google DeepMind AI Masters Scholarship application form at https://deepmind.google/education/
- [ ] Upload motivation letter (this document, 300-500 words)
- [ ] Upload research statement (this document, 400-600 words)
- [ ] Upload essay on leadership and AI for social good in Africa (this document, 200-350 words)
- [ ] Obtain and upload academic transcripts from University of Ibadan (B.Pharm, CGPA 5.1/7.0)
- [ ] Obtain and upload proof of enrollment or admission to M.Sc. Digital Health at HPI/Potsdam for Winter Semester 2026/27
- [ ] Prepare CV listing employment history (Synthcare, Ramset Pharmacy, CDDDP, GHRU-GSAR) and research lines
- [ ] Prepare list of preprints with URLs (OSF/Zenodo for CCT, TOPOLOGIX, cardiotoxicity study, interface-topology study)
- [ ] Prepare ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) links
- [ ] Prepare two reference letters, ideally from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar
- [ ] Verify deadline: 2025-04-30
- [ ] Confirm whether IIE administers this scholarship and whether additional eligibility criteria apply

EDITOR NOTES

- Eligibility risk: The profile lists enrollment in M.Sc. Digital Health at HPI/Potsdam for Winter Semester 2026/27, but the scholarship deadline is 2025-04-30. Verify whether the scholarship requires current enrollment or admission for the upcoming academic year, and whether the program start date aligns with the scholarship's funding period. If the M.Sc. start date is after the scholarship's funding window, this is a critical mismatch that must be resolved before submission.

- Verification needed: Confirm that HPI/Potsdam is a partner university for the DeepMind scholarship. The profile states the program is at a partner-eligible university, but this must be verified against the official partner list. If HPI is not a partner, the application is ineligible regardless of merit.

- Gap to fill: The essay on leadership and AI for social good in Africa is 200-350 words, but the draft above is longer. The applicant must trim it to the required limit. Also, the applicant should insert a specific anecdote or example from their clinical pharmacy work in Nigeria that illustrates the antimicrobial resistance problem, as the current draft references it generally but lacks a concrete patient or clinical scenario.

- Fact check: The profile states TOPOLOGIX beats mCSM-lig at approximately 0.70 AUROC, but the exact comparison methodology should be verified. The applicant should confirm whether mCSM-lig was run on the same Platinum benchmark or a different dataset, as this affects the validity of the comparison claim.

- Honest framing: The cardiotoxicity and interface-topology studies are negative results. The draft presents them as evidence of rigor, which is appropriate, but the applicant must ensure they are not described as current, ongoing work. They are completed, published preprints that motivated the pivot to TOPOLOGIX.

AI Draft — Google DeepMind Artificial Intelligence Masters Scholarship for African Students