Draft — Google DeepMind Scholarships to study Masters in ML & AI at Stellenbosch University

MOTIVATION LETTER

The Platinum benchmark contains 553 drug-resistance mutations. Structure-based tools like mCSM-lig can only score about 18 percent of them because most lack a resolved crystal structure. My TOPOLOGIX pipeline, built on ESM-2 protein language model delta-embeddings and Morgan fingerprints with a Random Forest classifier, covers 100 percent of those mutations and reaches an AUROC of 0.804 plus or minus 0.025. That result came from independent work completed without a formal research position, and it is the reason I am applying to the Google DeepMind Scholarships to study Masters in ML and AI at Stellenbosch University.

I am a Nigerian pharmacist and independent computational researcher. My undergraduate degree is from the University of Ibadan, where I graduated with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9. I am currently enrolled in the M.Sc. Digital Health programme at the Hasso Plattner Institute and the University of Potsdam, starting in the Winter Semester 2026/27. The Stellenbosch MSc in Machine Learning and Artificial Intelligence offers something my current programme does not: a full year of rigorous, formal training in the theoretical foundations of ML, from optimization and probabilistic modeling to deep learning architectures. I have taught myself these subjects through project work, and I have reached the limit of what self-study can provide.

The DeepMind scholarship's mission to empower African AI talent matches my own trajectory. I have built and validated four independent DuckDB-based data pipelines across life sciences, tech, and social science domains. I self-host local LLM serving with llama.cpp and manage production Linux systems with CI/CD and automated backups. These are practical skills, but they were learned opportunistically. The Stellenbosch programme would replace that patchwork with a structured curriculum, and it would place me in a South African research environment where I can build collaborations across the continent rather than working alone from Nigeria.

My research goal for the MSc is to extend TOPOLOGIX from drug-resistance prediction to the harder problem of resistance emergence over time. The current model predicts whether a given mutation confers resistance. The next version should predict which mutations are likely to arise under selective pressure from a specific drug, which requires modeling the fitness landscape itself. That is a sequence-to-sequence problem with sparse labels, and it needs the kind of probabilistic deep learning and representation learning that the Stellenbosch curriculum covers directly. The work has a clear African application: antimicrobial resistance surveillance in Nigeria and across the continent currently relies on genomic pipelines like the one I helped build at GHRU-GSAR, but those pipelines identify resistance after it appears. A model that anticipates resistance could change how treatment protocols are designed.

I am applying to Stellenbosch because the programme's combination of formal ML training, African location, and DeepMind's explicit commitment to developing African research talent is the fastest path from where I am to where my research needs to go.

STATEMENT OF RESEARCH INTEREST

My research interest sits at the intersection of protein machine learning and pharmacological systems modeling. I build computational tools that predict how drugs and biological systems interact, and I validate those tools against pre-registered benchmarks rather than retrospective convenience. The Stellenbosch MSc in Machine Learning and Artificial Intelligence is the formal training ground I need to take this work from independent projects to a sustainable research programme.

The core project I will bring into the MSc is TOPOLOGIX. It 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, it 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, which scores around 0.70, while covering every mutation in the benchmark. Structure-based tools cover only about 18 percent of mutations because most lack resolved crystal structures. This coverage gap is the practical problem I want to solve: resistance prediction should not depend on whether a structure happens to exist.

TOPOLOGIX did not start as a sequence model. It started as a topology project. I tested whether bipartite persistent homology, using an opposition-distance metric implemented with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. In a pre-registered, powered replication, the topological features did not beat a plain descriptor baseline, with an AUROC of 0.8426 versus 0.8782. I then applied the same topological constructs to drug-resistance prediction and found they carried almost no signal, with AUROCs of 0.425 and 0.485 on the Platinum benchmark. Those negative results are published as preprints and were the direct motivation for switching to sequence representations. The lesson I took from that arc is that method choice must follow the data, not the fashion.

The next phase of this research, which I intend to pursue as my MSc thesis, is predicting resistance emergence rather than resistance presence. The current TOPOLOGIX model answers a classification question: does this mutation confer resistance? The harder question is generative: given a drug and a starting viral or bacterial sequence, which mutations are most likely to arise under selective pressure? This requires modeling the fitness landscape as a function of sequence and drug, which in turn requires variational inference, normalizing flows, or related generative architectures. These are exactly the methods taught in the Stellenbosch curriculum, and they are methods I have not yet had the opportunity to study formally.

The African relevance is direct. Antimicrobial resistance is a documented crisis across the continent, and genomic surveillance pipelines like the one I contributed to at GHRU-GSAR in Nigeria identify resistance only after it has emerged in a patient. A model that anticipates resistance could inform treatment guidelines before clinical failure occurs. The Stellenbosch programme, with its DeepMind scholarship support, would give me the training, the computing resources, and the research community to build that model properly.

I am also bringing a secondary line of work into the programme. My neurocascade project is a receptor-to-behavior brain-circuit simulation engine that couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics, with Bayesian calibration via PyMC. All 62 tests pass, and the circuit-layer parameters are explicitly labeled illustrative pending real behavioral data. This project is methodologically adjacent to TOPOLOGIX in that both depend on careful probabilistic modeling, but it is not the focus of my MSc. The focus is TOPOLOGIX and resistance emergence.

The Stellenbosch programme does not require me to identify a supervisor at application, which suits my situation. I have worked independently for two years, and I have the self-direction to define a thesis project. What I need from the programme is the theoretical depth and the collaborative environment to execute it at a higher level than I can achieve alone.

EDITOR NOTES

- Framing choice: I selected TOPOLOGIX as the anchor research line because it is the only project in the profile that directly demonstrates existing ML expertise (ESM-2, Random Forest, benchmark evaluation) applied to a biomedical problem, which matches the programme's ML/AI focus. The CCT model is more pharmacologically sophisticated but is an ODE/Bayesian project with no deep learning component, making it a weaker fit for an ML MSc. The negative topology results are included as honest provenance for TOPOLOGIX, not as current work.

- Eligibility risk: The profile states Eniola is Nigerian, which satisfies the African citizenship requirement for the Google DeepMind scholarship. However, the scholarship is administered by IIE, and the profile does not confirm whether IIE's criteria include financial need or leadership potential. The applicant should verify IIE's specific criteria before submitting.

- Deadline conflict: The scholarship consideration deadline is 15 August 2026, and general admission closes 31 October 2026. The profile lists the programme deadline as 2026-12-31, which may be a third date. The applicant must confirm which deadline applies to the scholarship track and submit well before August if possible.

- Verification needed: The profile states the applicant is enrolled in M.Sc. Digital Health at HPI/Potsdam starting Winter Semester 2026/27. The motivation letter references this. The applicant should confirm that enrolling in a second MSc concurrently is permitted by both institutions and that Stellenbosch will accept an applicant already enrolled elsewhere.

- Gap to fill: The motivation letter does not mention any specific Stellenbosch faculty member or research group. The programme does not require supervisor identification, but the applicant should research the ML/AI group's current projects and name one or two relevant faculty in the final version if possible, to strengthen the fit argument.

- Gap to fill: The statement of research interest proposes extending TOPOLOGIX to resistance emergence prediction. This is a plausible next step, but the applicant should verify that the Platinum benchmark or a related dataset contains longitudinal or selective-pressure data to support this extension. If not, the thesis project description may need adjustment.

- Tone check: The motivation letter opens with a data point rather than self-introduction, which follows the formatting rules. The phrase "I am not applying to Stellenbosch because it is convenient" is a deliberate contrast to the throat-clearing openers the rules prohibit. The applicant should ensure this tone is consistent with any additional materials Stellenbosch requires beyond the three documents listed.

- Document count: The submission page requires exactly three documents: a thorough CV, a letter of motivation, and a statement of research interest. This draft provides the letter and statement. The CV must be compiled separately from the profile data, and the applicant should ensure it includes the ORCID, GitHub, and personal site URLs listed in the profile.

CHECKLIST

- [ ] Verify IIE Google DeepMind scholarship eligibility criteria, specifically financial need and leadership potential requirements
- [ ] Confirm the exact scholarship application deadline (15 August 2026 vs 31 October 2026 vs 2026-12-31) on the SUNStudent portal
- [ ] Confirm that concurrent enrollment in M.Sc. Digital Health at HPI/Potsdam and M.Sc. ML/AI at Stellenbosch is permitted by both institutions
- [ ] Research Stellenbosch ML/AI faculty and identify one or two relevant researchers to reference in the motivation letter
- [ ] Verify that the Platinum benchmark or a related dataset supports the proposed resistance emergence extension for the thesis project
- [ ] Compile thorough CV from profile data, including ORCID, GitHub, and zyco.org URLs
- [ ] Create SUNStudent applicant portal profile and log in
- [ ] Select Faculty of Science, Postgraduate, MSc (Machine Learning and Artificial Intelligence), full-time or part-time
- [ ] Complete all required information in the left panel menus
- [ ] Upload thorough CV
- [ ] Upload letter of motivation
- [ ] Upload statement of research interest
- [ ] Review and submit application
- [ ] Pay any applicable application fee
- [ ] Confirm receipt of application and scholarship nomination status with IIE

MOTIVATION LETTER

The Platinum benchmark contains 553 drug-resistance mutations. Structure-based tools like mCSM-lig can only score about 18 percent of them because most lack a resolved crystal structure. My TOPOLOGIX pipeline, built on ESM-2 protein language model delta-embeddings and Morgan fingerprints with a Random Forest classifier, covers 100 percent of those mutations and reaches an AUROC of 0.804 plus or minus 0.025. That result came from independent work completed without a formal research position, and it is the reason I am applying to the Google DeepMind Scholarships to study Masters in ML and AI at Stellenbosch University.

I am a Nigerian pharmacist and independent computational researcher. My undergraduate degree is from the University of Ibadan, where I graduated with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9. I am currently enrolled in the M.Sc. Digital Health programme at the Hasso Plattner Institute and the University of Potsdam, starting in the Winter Semester 2026/27. The Stellenbosch MSc in Machine Learning and Artificial Intelligence offers something my current programme does not: a full year of rigorous, formal training in the theoretical foundations of ML, from optimization and probabilistic modeling to deep learning architectures. I have taught myself these subjects through project work, and I have reached the limit of what self-study can provide.

The DeepMind scholarship's mission to empower African AI talent matches my own trajectory. I have built and validated four independent DuckDB-based data pipelines across life sciences, tech, and social science domains. I self-host local LLM serving with llama.cpp and manage production Linux systems with CI/CD and automated backups. These are practical skills, but they were learned opportunistically. The Stellenbosch programme would replace that patchwork with a structured curriculum, and it would place me in a South African research environment where I can build collaborations across the continent rather than working alone from Nigeria.

My research goal for the MSc is to extend TOPOLOGIX from drug-resistance prediction to the harder problem of resistance emergence over time. The current model predicts whether a given mutation confers resistance. The next version should predict which mutations are likely to arise under selective pressure from a specific drug, which requires modeling the fitness landscape itself. That is a sequence-to-sequence problem with sparse labels, and it needs the kind of probabilistic deep learning and representation learning that the Stellenbosch curriculum covers directly. The work has a clear African application: antimicrobial resistance surveillance in Nigeria and across the continent currently relies on genomic pipelines like the one I helped build at GHRU-GSAR, but those pipelines identify resistance after it appears. A model that anticipates resistance could change how treatment protocols are designed.

I am applying to Stellenbosch because the programme's combination of formal ML training, African location, and DeepMind's explicit commitment to developing African research talent is the fastest path from where I am to where my research needs to go.

STATEMENT OF RESEARCH INTEREST

My research interest sits at the intersection of protein machine learning and pharmacological systems modeling. I build computational tools that predict how drugs and biological systems interact, and I validate those tools against pre-registered benchmarks rather than retrospective convenience. The Stellenbosch MSc in Machine Learning and Artificial Intelligence is the formal training ground I need to take this work from independent projects to a sustainable research programme.

The core project I will bring into the MSc is TOPOLOGIX. It 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, it 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, which scores around 0.70, while covering every mutation in the benchmark. Structure-based tools cover only about 18 percent of mutations because most lack resolved crystal structures. This coverage gap is the practical problem I want to solve: resistance prediction should not depend on whether a structure happens to exist.

TOPOLOGIX did not start as a sequence model. It started as a topology project. I tested whether bipartite persistent homology, using an opposition-distance metric implemented with Ripser and GUDHI, could predict hERG cardiotoxicity from protein-ligand interface geometry. In a pre-registered, powered replication, the topological features did not beat a plain descriptor baseline, with an AUROC of 0.8426 versus 0.8782. I then applied the same topological constructs to drug-resistance prediction and found they carried almost no signal, with AUROCs of 0.425 and 0.485 on the Platinum benchmark. Those negative results are published as preprints and were the direct motivation for switching to sequence representations. The lesson I took from that arc is that method choice must follow the data, not the fashion.

The next phase of this research, which I intend to pursue as my MSc thesis, is predicting resistance emergence rather than resistance presence. The current TOPOLOGIX model answers a classification question: does this mutation confer resistance? The harder question is generative: given a drug and a starting viral or bacterial sequence, which mutations are most likely to arise under selective pressure? This requires modeling the fitness landscape as a function of sequence and drug, which in turn requires variational inference, normalizing flows, or related generative architectures. These are exactly the methods taught in the Stellenbosch curriculum, and they are methods I have not yet had the opportunity to study formally.

The African relevance is direct. Antimicrobial resistance is a documented crisis across the continent, and genomic surveillance pipelines like the one I contributed to at GHRU-GSAR in Nigeria identify resistance only after it has emerged in a patient. A model that anticipates resistance could inform treatment guidelines before clinical failure occurs. The Stellenbosch programme, with its DeepMind scholarship support, would give me the training, the computing resources, and the research community to build that model properly.

I am also bringing a secondary line of work into the programme. My neurocascade project is a receptor-to-behavior brain-circuit simulation engine that couples pharmacokinetics to receptor binding to Wilson-Cowan circuit dynamics, with Bayesian calibration via PyMC. All 62 tests pass, and the circuit-layer parameters are explicitly labeled illustrative pending real behavioral data. This project is methodologically adjacent to TOPOLOGIX in that both depend on careful probabilistic modeling, but it is not the focus of my MSc. The focus is TOPOLOGIX and resistance emergence.

The Stellenbosch programme does not require me to identify a supervisor at application, which suits my situation. I have worked independently for two years, and I have the self-direction to define a thesis project. What I need from the programme is the theoretical depth and the collaborative environment to execute it at a higher level than I can achieve alone.

EDITOR NOTES

- Framing choice: I selected TOPOLOGIX as the anchor research line because it is the only project in the profile that directly demonstrates existing ML expertise (ESM-2, Random Forest, benchmark evaluation) applied to a biomedical problem, which matches the programme's ML/AI focus. The CCT model is more pharmacologically sophisticated but is an ODE/Bayesian project with no deep learning component, making it a weaker fit for an ML MSc. The negative topology results are included as honest provenance for TOPOLOGIX, not as current work.

- Eligibility risk: The profile states Eniola is Nigerian, which satisfies the African citizenship requirement for the Google DeepMind scholarship. However, the scholarship is administered by IIE, and the profile does not confirm whether IIE's criteria include financial need or leadership potential. The applicant should verify IIE's specific criteria before submitting.

- Deadline conflict: The scholarship consideration deadline is 15 August 2026, and general admission closes 31 October 2026. The profile lists the programme deadline as 2026-12-31, which may be a third date. The applicant must confirm which deadline applies to the scholarship track and submit well before August if possible.

- Verification needed: The profile states the applicant is enrolled in M.Sc. Digital Health at HPI/Potsdam starting Winter Semester 2026/27. The motivation letter references this. The applicant should confirm that enrolling in a second MSc concurrently is permitted by both institutions and that Stellenbosch will accept an applicant already enrolled elsewhere.

- Gap to fill: The motivation letter does not mention any specific Stellenbosch faculty member or research group. The programme does not require supervisor identification, but the applicant should research the ML/AI group's current projects and name one or two relevant faculty in the final version if possible, to strengthen the fit argument.

- Gap to fill: The statement of research interest proposes extending TOPOLOGIX to resistance emergence prediction. This is a plausible next step, but the applicant should verify that the Platinum benchmark or a related dataset contains longitudinal or selective-pressure data to support this extension. If not, the thesis project description may need adjustment.

- Tone check: The motivation letter opens with a data point rather than self-introduction, which follows the formatting rules. The phrase "I am not applying to Stellenbosch because it is convenient" is a deliberate contrast to the throat-clearing openers the rules prohibit. The applicant should ensure this tone is consistent with any additional materials Stellenbosch requires beyond the three documents listed.

- Document count: The submission page requires exactly three documents: a thorough CV, a letter of motivation, and a statement of research interest. This draft provides the letter and statement. The CV must be compiled separately from the profile data, and the applicant should ensure it includes the ORCID, GitHub, and personal site URLs listed in the profile.

CHECKLIST

- [ ] Verify IIE Google DeepMind scholarship eligibility criteria, specifically financial need and leadership potential requirements
- [ ] Confirm the exact scholarship application deadline (15 August 2026 vs 31 October 2026 vs 2026-12-31) on the SUNStudent portal
- [ ] Confirm that concurrent enrollment in M.Sc. Digital Health at HPI/Potsdam and M.Sc. ML/AI at Stellenbosch is permitted by both institutions
- [ ] Research Stellenbosch ML/AI faculty and identify one or two relevant researchers to reference in the motivation letter
- [ ] Verify that the Platinum benchmark or a related dataset supports the proposed resistance emergence extension for the thesis project
- [ ] Compile thorough CV from profile data, including ORCID, GitHub, and zyco.org URLs
- [ ] Create SUNStudent applicant portal profile and log in
- [ ] Select Faculty of Science, Postgraduate, MSc (Machine Learning and Artificial Intelligence), full-time or part-time
- [ ] Complete all required information in the left panel menus
- [ ] Upload thorough CV
- [ ] Upload letter of motivation
- [ ] Upload statement of research interest
- [ ] Review and submit application
- [ ] Pay any applicable application fee
- [ ] Confirm receipt of application and scholarship nomination status with IIE

AI Draft — Google DeepMind Scholarships to study Masters in ML & AI at Stellenbosch University