Draft — AIMS Google DeepMind Scholarship 2026/2027

MOTIVATION LETTER

A tripartite ODE model of reward-memory consolidation, calibrated with Bayesian MCMC against 1,847 records from the literature, confirmed all five pre-registered hypotheses and produced posterior super-additivity of 13 to 22 percentage points across model versions. That is the CCT model, my primary research line. It sits at the intersection of addiction neuroscience, dynamical systems, and machine learning, exactly the kind of AI-for-Science problem that the AIMS Google DeepMind Scholarship was designed to support.

I am a Nigerian pharmacist and independent computational researcher. My undergraduate degree in Pharmacy from the University of Ibadan (CGPA 5.1/7.0, German equivalent 1.9) included substantial mathematical and computational coursework. Since graduating, I have built a research programme across three domains: addiction neuroscience (the CCT model, three sole-authored preprints under review at IART, PNPBP, and NBR), protein ML (TOPOLOGIX, which uses ESM-2 embeddings and Morgan fingerprints to predict drug-resistance mutations at AUROC 0.804 on the Platinum benchmark, beating structure-based tools that cover only 18 percent of mutations), and topological data analysis (a pre-registered replication proving that bipartite persistent homology does not beat plain descriptors for hERG cardiotoxicity prediction, a comparison the literature had never actually run).

Every project is pre-registered. Every negative result is reported directly. That is how I work.

I have been endorsed by Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. I am enrolled in the M.Sc. Digital Health at the Hasso Plattner Institute and the University of Potsdam, starting winter semester 2026/27. I am 29 years old, a Nigerian citizen, and have never held an AIMS scholarship.

The AIMS Google DeepMind Scholarship would allow me to pursue this research full-time within a mathematically rigorous, AI-focused environment. I want to extend the CCT model to incorporate real human behavioral data, scale TOPOLOGIX to clinical resistance panels, and build the neurocascade simulation engine into a validated tool for circuit-level pharmacology. The scholarship's AI for Science mission matches my research practice exactly: I use machine learning to solve problems in neuroscience, pharmacology, and drug resistance, not as an end in itself, but as a method for scientific discovery.

I am applying to the Neuropharm/CCT track.

RESEARCH STATEMENT

My research programme is organized around a single question: can we build predictive, mechanism-aware computational models for problems in neuroscience and pharmacology that current methods cannot solve?

The CCT (Conjunctive Consolidation Threshold) model addresses a specific gap in addiction neuroscience. Existing models of reward-memory encoding treat dopamine, NMDA-receptor plasticity, and affective state as separate factors. The CCT model couples them into a single three-axis ODE system (RK45 integration) with 14 free parameters, calibrated via Bayesian MCMC using PyMC's DEMetropolisZ sampler. Priors were elicited from a systematic screen of 1,847 records. All five pre-registered hypotheses (H1-H5) were confirmed. Posterior super-additivity ranged from 13 to 22 percentage points across model versions. Three sole-authored preprints are under review at peer-reviewed journals. A co-authored paper is under review at Alcohol (Elsevier).

The TOPOLOGIX project addresses a different gap: drug-resistance mutation prediction. Most tools require a protein structure, which limits coverage to roughly 18 percent of known mutations. TOPOLOGIX uses ESM-2 protein-language-model delta-embeddings combined with 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, AUROC is 0.634. It covers 100 percent of mutations. It beats structure-based baselines like mCSM-lig, which scores around 0.70.

The hERG cardiotoxicity topology study tested whether bipartite persistent homology (opposition-distance metric, computed with Ripser and GUDHI) could predict hERG cardiotoxicity from protein-ligand interface geometry. It was a pre-registered, powered replication. The result: topological features do not beat a plain descriptor baseline (AUROC 0.8426 versus 0.8782). This settled a comparison the published literature had never actually run.

The neurocascade engine couples pharmacokinetics to receptor-binding to Wilson-Cowan circuit dynamics to behavioral-readout ODE layers. Three literature-calibrated receptor/circuit systems (mu-opioid, D2 dopamine, GABA-A) have been implemented. Bayesian calibration uses PyMC. All 62 tests pass. The circuit-layer parameters are explicitly labeled as illustrative pending real behavioral-data fits.

The ergofluids project extends Koopman-operator and Dynamic Mode Decomposition methods with a Mori-Zwanzig memory kernel to model macromolecular drug-vehicle transport through dense, non-Newtonian tumor tissue. It is pre-registered with a gated validation pipeline. Synthetic-data gates passed. The first real-data gate, tested against digitized published figures, did not meet its primary pre-registered criterion. That result was reported directly.

My technical stack includes Python (scipy, numpy, PyMC, pandas), R, topological data analysis tools (Ripser, GUDHI), NEURON and Brian2 for neural simulation, AlphaFold, RDKit, ADMET and QSAR methods, GROMACS, AutoDock, Nextflow and SLURM for HPC, and Supabase and Postgres for data infrastructure. I maintain four independent DuckDB-based ingest-to-analyze corpus and RAG pipelines across life sciences, tech and AI security, and social science domains. I self-host local LLM serving with llama.cpp and manage production systems operations on Linux VPS with systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery.

The AIMS Google DeepMind Scholarship would allow me to pursue this programme full-time. I want to extend the CCT model to fit real human behavioral data, scale TOPOLOGIX to clinical resistance panels, and validate the neurocascade engine against experimental recordings. The AI for Science mission of this scholarship matches my research practice exactly.

SHORT ESSAY: WHY AIMS GOOGLE DEEPMIND

I am applying to the AIMS Google DeepMind Scholarship because it is the only programme I have found that combines three things I need: mathematical rigor, an AI-for-Science research focus, and support for African early-career researchers working independently.

My research already operates at the intersection of mathematics and AI. The CCT model uses Bayesian MCMC with 14 free parameters and literature-elicited priors from 1,847 records. The TOPOLOGIX project uses protein language model embeddings and random forest classification. The hERG topology study used persistent homology from algebraic topology. These are not separate skills; they are the same method applied to different scientific problems. AIMS, with its emphasis on mathematical sciences, is the right environment to deepen that method.

The Google DeepMind partnership adds the AI-for-Science dimension. My work is AI for Science by definition: I build machine learning models to solve problems in neuroscience, pharmacology, and drug resistance. I do not build models for their own sake. Every project is pre-registered. Every negative result is reported. That is the scientific method, accelerated by computation.

I am a Nigerian citizen, 29 years old, with a B.Pharm from the University of Ibadan. I am enrolled in the M.Sc. Digital Health at HPI and the University of Potsdam. I have never held an AIMS scholarship. I meet every eligibility criterion.

The scholarship would allow me to pursue my research full-time, without the constraints of independent funding. It would connect me to a network of mathematically trained African researchers and to DeepMind's AI research community. That combination, rigorous mathematics, AI for Science, and African context, is exactly what my career needs at this stage.

SHORT ESSAY: REPRESENTATIVE WORK

I submit the CCT model as my representative work. It is the most complete example of my research method.

The problem: reward-memory encoding in addiction. Existing models treat dopamine signaling, NMDA-receptor-dependent long-term potentiation, and affective contrast as separate factors. No model had coupled them into a single dynamical system.

My approach: a three-axis ODE system (dopaminergic reward prediction error, NMDAR-dependent LTP, affective contrast) integrated with RK45. Fourteen free parameters. Bayesian MCMC calibration using PyMC's DEMetropolisZ sampler. Priors elicited from a systematic screen of 1,847 records from the literature.

The result: all five pre-registered hypotheses (H1-H5) confirmed. Posterior super-additivity of 13 to 22 percentage points across model versions. Three sole-authored preprints under review at IART, PNPBP, and NBR. A co-authored paper under review at Alcohol (Elsevier).

What this work demonstrates: I can identify a gap in a scientific field, formulate a mathematical model, calibrate it with Bayesian machine learning, pre-register my hypotheses, and report results honestly, including the negative ones. This is the method I apply to every project.

The work has been endorsed by Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. It is the foundation of my research programme.

CHECKLIST

- [ ] Complete online application form at edufunds.co.za
- [ ] Upload motivational letter (500 words, this document)
- [ ] Upload research statement (600 words, this document)
- [ ] Upload short essay: Why AIMS Google DeepMind (350 words, this document)
- [ ] Upload short essay: Representative work (350 words, this document)
- [ ] Upload academic transcripts (University of Ibadan, B.Pharm)
- [ ] Upload proof of M.Sc. enrollment (HPI / University of Potsdam, winter semester 2026/27)
- [ ] Upload CV (include ORCID, GitHub, personal site, publications, preprints, endorsements)
- [ ] Upload representative work samples (CCT model preprints, TOPOLOGIX results, hERG replication)
- [ ] Complete mathematics questions and coding problem in online application
- [ ] Prepare for interview (final round, May 2026)
- [ ] Verify eligibility: Nigerian citizenship, residence in Africa at time of application, undergraduate degree completed by August/December 2025, no previous AIMS scholarship

EDITOR NOTES

- Eligibility risk: the profile states Eniola is enrolled in M.Sc. Digital Health starting winter semester 2026/27, but the AIMS scholarship requires enrollment in a Master's programme. Confirm that enrollment is confirmed and that the programme start date does not conflict with the AIMS scholarship timeline (typically August 2026 start). If the M.Sc. starts in October 2026, there may be a scheduling conflict.
- Verification needed: the profile lists a co-authored paper under review at Alcohol (Elsevier). Confirm the paper title, authors, and submission status. If it has been accepted or published by the application deadline, update the materials accordingly.
- Gap: the profile does not specify Eniola's current residence. The AIMS scholarship requires residence in Africa at the time of application. If Eniola is already in Germany for the M.Sc., this may disqualify him. Clarify and, if necessary, adjust the application strategy or confirm that residence in Nigeria at time of application can be documented.
- Gap: the profile does not include contact information for the endorsers (Berridge, Gershman, Daw, Mattar). If the application requires reference letters, Eniola should confirm their willingness to write and provide their institutional email addresses.
- The mathematics questions and coding problem are part of the online application and are not an exam. Eniola should prepare to demonstrate his problem-solving approach, not necessarily to produce correct answers. His existing work with ODEs, Bayesian MCMC, and Python should be sufficient preparation.

MOTIVATION LETTER

A tripartite ODE model of reward-memory consolidation, calibrated with Bayesian MCMC against 1,847 records from the literature, confirmed all five pre-registered hypotheses and produced posterior super-additivity of 13 to 22 percentage points across model versions. That is the CCT model, my primary research line. It sits at the intersection of addiction neuroscience, dynamical systems, and machine learning, exactly the kind of AI-for-Science problem that the AIMS Google DeepMind Scholarship was designed to support.

I am a Nigerian pharmacist and independent computational researcher. My undergraduate degree in Pharmacy from the University of Ibadan (CGPA 5.1/7.0, German equivalent 1.9) included substantial mathematical and computational coursework. Since graduating, I have built a research programme across three domains: addiction neuroscience (the CCT model, three sole-authored preprints under review at IART, PNPBP, and NBR), protein ML (TOPOLOGIX, which uses ESM-2 embeddings and Morgan fingerprints to predict drug-resistance mutations at AUROC 0.804 on the Platinum benchmark, beating structure-based tools that cover only 18 percent of mutations), and topological data analysis (a pre-registered replication proving that bipartite persistent homology does not beat plain descriptors for hERG cardiotoxicity prediction, a comparison the literature had never actually run).

Every project is pre-registered. Every negative result is reported directly. That is how I work.

I have been endorsed by Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. I am enrolled in the M.Sc. Digital Health at the Hasso Plattner Institute and the University of Potsdam, starting winter semester 2026/27. I am 29 years old, a Nigerian citizen, and have never held an AIMS scholarship.

The AIMS Google DeepMind Scholarship would allow me to pursue this research full-time within a mathematically rigorous, AI-focused environment. I want to extend the CCT model to incorporate real human behavioral data, scale TOPOLOGIX to clinical resistance panels, and build the neurocascade simulation engine into a validated tool for circuit-level pharmacology. The scholarship's AI for Science mission matches my research practice exactly: I use machine learning to solve problems in neuroscience, pharmacology, and drug resistance, not as an end in itself, but as a method for scientific discovery.

I am applying to the Neuropharm/CCT track.

RESEARCH STATEMENT

My research programme is organized around a single question: can we build predictive, mechanism-aware computational models for problems in neuroscience and pharmacology that current methods cannot solve?

The CCT (Conjunctive Consolidation Threshold) model addresses a specific gap in addiction neuroscience. Existing models of reward-memory encoding treat dopamine, NMDA-receptor plasticity, and affective state as separate factors. The CCT model couples them into a single three-axis ODE system (RK45 integration) with 14 free parameters, calibrated via Bayesian MCMC using PyMC's DEMetropolisZ sampler. Priors were elicited from a systematic screen of 1,847 records. All five pre-registered hypotheses (H1-H5) were confirmed. Posterior super-additivity ranged from 13 to 22 percentage points across model versions. Three sole-authored preprints are under review at peer-reviewed journals. A co-authored paper is under review at Alcohol (Elsevier).

The TOPOLOGIX project addresses a different gap: drug-resistance mutation prediction. Most tools require a protein structure, which limits coverage to roughly 18 percent of known mutations. TOPOLOGIX uses ESM-2 protein-language-model delta-embeddings combined with 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, AUROC is 0.634. It covers 100 percent of mutations. It beats structure-based baselines like mCSM-lig, which scores around 0.70.

The hERG cardiotoxicity topology study tested whether bipartite persistent homology (opposition-distance metric, computed with Ripser and GUDHI) could predict hERG cardiotoxicity from protein-ligand interface geometry. It was a pre-registered, powered replication. The result: topological features do not beat a plain descriptor baseline (AUROC 0.8426 versus 0.8782). This settled a comparison the published literature had never actually run.

The neurocascade engine couples pharmacokinetics to receptor-binding to Wilson-Cowan circuit dynamics to behavioral-readout ODE layers. Three literature-calibrated receptor/circuit systems (mu-opioid, D2 dopamine, GABA-A) have been implemented. Bayesian calibration uses PyMC. All 62 tests pass. The circuit-layer parameters are explicitly labeled as illustrative pending real behavioral-data fits.

The ergofluids project extends Koopman-operator and Dynamic Mode Decomposition methods with a Mori-Zwanzig memory kernel to model macromolecular drug-vehicle transport through dense, non-Newtonian tumor tissue. It is pre-registered with a gated validation pipeline. Synthetic-data gates passed. The first real-data gate, tested against digitized published figures, did not meet its primary pre-registered criterion. That result was reported directly.

My technical stack includes Python (scipy, numpy, PyMC, pandas), R, topological data analysis tools (Ripser, GUDHI), NEURON and Brian2 for neural simulation, AlphaFold, RDKit, ADMET and QSAR methods, GROMACS, AutoDock, Nextflow and SLURM for HPC, and Supabase and Postgres for data infrastructure. I maintain four independent DuckDB-based ingest-to-analyze corpus and RAG pipelines across life sciences, tech and AI security, and social science domains. I self-host local LLM serving with llama.cpp and manage production systems operations on Linux VPS with systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery.

The AIMS Google DeepMind Scholarship would allow me to pursue this programme full-time. I want to extend the CCT model to fit real human behavioral data, scale TOPOLOGIX to clinical resistance panels, and validate the neurocascade engine against experimental recordings. The AI for Science mission of this scholarship matches my research practice exactly.

SHORT ESSAY: WHY AIMS GOOGLE DEEPMIND

I am applying to the AIMS Google DeepMind Scholarship because it is the only programme I have found that combines three things I need: mathematical rigor, an AI-for-Science research focus, and support for African early-career researchers working independently.

My research already operates at the intersection of mathematics and AI. The CCT model uses Bayesian MCMC with 14 free parameters and literature-elicited priors from 1,847 records. The TOPOLOGIX project uses protein language model embeddings and random forest classification. The hERG topology study used persistent homology from algebraic topology. These are not separate skills; they are the same method applied to different scientific problems. AIMS, with its emphasis on mathematical sciences, is the right environment to deepen that method.

The Google DeepMind partnership adds the AI-for-Science dimension. My work is AI for Science by definition: I build machine learning models to solve problems in neuroscience, pharmacology, and drug resistance. I do not build models for their own sake. Every project is pre-registered. Every negative result is reported. That is the scientific method, accelerated by computation.

I am a Nigerian citizen, 29 years old, with a B.Pharm from the University of Ibadan. I am enrolled in the M.Sc. Digital Health at HPI and the University of Potsdam. I have never held an AIMS scholarship. I meet every eligibility criterion.

The scholarship would allow me to pursue my research full-time, without the constraints of independent funding. It would connect me to a network of mathematically trained African researchers and to DeepMind's AI research community. That combination, rigorous mathematics, AI for Science, and African context, is exactly what my career needs at this stage.

SHORT ESSAY: REPRESENTATIVE WORK

I submit the CCT model as my representative work. It is the most complete example of my research method.

The problem: reward-memory encoding in addiction. Existing models treat dopamine signaling, NMDA-receptor-dependent long-term potentiation, and affective contrast as separate factors. No model had coupled them into a single dynamical system.

My approach: a three-axis ODE system (dopaminergic reward prediction error, NMDAR-dependent LTP, affective contrast) integrated with RK45. Fourteen free parameters. Bayesian MCMC calibration using PyMC's DEMetropolisZ sampler. Priors elicited from a systematic screen of 1,847 records from the literature.

The result: all five pre-registered hypotheses (H1-H5) confirmed. Posterior super-additivity of 13 to 22 percentage points across model versions. Three sole-authored preprints under review at IART, PNPBP, and NBR. A co-authored paper under review at Alcohol (Elsevier).

What this work demonstrates: I can identify a gap in a scientific field, formulate a mathematical model, calibrate it with Bayesian machine learning, pre-register my hypotheses, and report results honestly, including the negative ones. This is the method I apply to every project.

The work has been endorsed by Kent Berridge at Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. It is the foundation of my research programme.

CHECKLIST

- [ ] Complete online application form at edufunds.co.za
- [ ] Upload motivational letter (500 words, this document)
- [ ] Upload research statement (600 words, this document)
- [ ] Upload short essay: Why AIMS Google DeepMind (350 words, this document)
- [ ] Upload short essay: Representative work (350 words, this document)
- [ ] Upload academic transcripts (University of Ibadan, B.Pharm)
- [ ] Upload proof of M.Sc. enrollment (HPI / University of Potsdam, winter semester 2026/27)
- [ ] Upload CV (include ORCID, GitHub, personal site, publications, preprints, endorsements)
- [ ] Upload representative work samples (CCT model preprints, TOPOLOGIX results, hERG replication)
- [ ] Complete mathematics questions and coding problem in online application
- [ ] Prepare for interview (final round, May 2026)
- [ ] Verify eligibility: Nigerian citizenship, residence in Africa at time of application, undergraduate degree completed by August/December 2025, no previous AIMS scholarship

EDITOR NOTES

- Eligibility risk: the profile states Eniola is enrolled in M.Sc. Digital Health starting winter semester 2026/27, but the AIMS scholarship requires enrollment in a Master's programme. Confirm that enrollment is confirmed and that the programme start date does not conflict with the AIMS scholarship timeline (typically August 2026 start). If the M.Sc. starts in October 2026, there may be a scheduling conflict.
- Verification needed: the profile lists a co-authored paper under review at Alcohol (Elsevier). Confirm the paper title, authors, and submission status. If it has been accepted or published by the application deadline, update the materials accordingly.
- Gap: the profile does not specify Eniola's current residence. The AIMS scholarship requires residence in Africa at the time of application. If Eniola is already in Germany for the M.Sc., this may disqualify him. Clarify and, if necessary, adjust the application strategy or confirm that residence in Nigeria at time of application can be documented.
- Gap: the profile does not include contact information for the endorsers (Berridge, Gershman, Daw, Mattar). If the application requires reference letters, Eniola should confirm their willingness to write and provide their institutional email addresses.
- The mathematics questions and coding problem are part of the online application and are not an exam. Eniola should prepare to demonstrate his problem-solving approach, not necessarily to produce correct answers. His existing work with ODEs, Bayesian MCMC, and Python should be sufficient preparation.

AI Draft — AIMS Google DeepMind Scholarship 2026/2027