Draft — AIMS Google DeepMind Scholarship 2026 – AI for Science

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, or CCT, is a Bayesian-calibrated dynamical system that predicts whether a reward-memory can be pharmacologically prevented from encoding. It couples dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast into a single three-axis ODE model, solved with RK45 and calibrated with PyMC DEMetropolisZ across fourteen free parameters. The priors come from a systematic screen of 1,847 records in the addiction neuroscience literature. All five pre-registered hypotheses, H1 through H5, were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. This is AI for science in the strictest sense: a mathematical model of a biological mechanism, fitted to empirical data, making falsifiable predictions about a disease that kills hundreds of thousands of people each year.

I am a Nigerian pharmacist and independent computational researcher, currently enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and the University of Potsdam. 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 have spent the last two years building the CCT model, the TOPOLOGIX protein-language-model pipeline for drug-resistance prediction, and a receptor-to-behavior simulation engine called neurocascade. The CCT work is currently under review at three peer-reviewed journals: International Addiction Review, Progress in Neuro-Psychopharmacology and Biological Psychiatry, and Neuroscience and Biobehavioral Reviews. A co-authored paper is under review at Alcohol, an Elsevier journal.

The AIMS Google DeepMind Scholarship is the correct next step for this work. AIMS exists to develop mathematical sciences talent in Africa, and DeepMind brings world-class machine learning research capacity. The CCT model is mathematically rigorous but methodologically conservative; it uses classical ODE solving and MCMC sampling. The mentorship and coursework available through this scholarship would let me apply modern machine learning methods, including neural ODEs and amortized variational inference, to scale the model from three receptor systems to the full mesolimbic circuitry. The bidirectional mentorship model, where I both learn from DeepMind researchers and contribute my domain knowledge in neuropharmacology, is precisely the structure that would accelerate this research.

I meet the eligibility criteria: I am a Nigerian citizen, I hold a four-year degree, and I have never held an AIMS scholarship. My academic background spans pharmacy, computational modeling, and software engineering, all of which are directly relevant to the AI for Science focus. The mathematics problem set and coding responses required by the application are tasks I am prepared to complete at a high level; my daily work involves solving ODE systems, implementing persistent homology pipelines, and building production-grade data infrastructure.

The CCT model makes a concrete clinical claim: that a combination of pharmacological agents acting on dopaminergic, glutamatergic, and affective systems can prevent the consolidation of reward memories, which is the core mechanism of addiction. If validated in clinical populations, this would change how relapse prevention is approached. The AIMS Google DeepMind Scholarship would give me the mathematical training and research network to move this work from preprint to clinical translation.

RESEARCH STATEMENT

The CCT model addresses a specific gap in addiction neuroscience: the field has abundant data on individual receptors and circuits, but no quantitative framework that predicts how pharmacological interventions across multiple systems interact to prevent reward-memory encoding. My contribution is a tripartite model that treats reward-memory consolidation as a threshold phenomenon, where the probability of encoding depends on the joint state of three coupled axes: dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast.

The model is a system of ordinary differential equations solved with RK45. It has fourteen free parameters, each with priors elicited from a systematic literature screen of 1,847 records. I calibrated the model using Bayesian MCMC with the DEMetropolisZ sampler in PyMC. The calibration converged, and all five pre-registered hypotheses were confirmed. The key finding is super-additivity: the combined effect of interventions across the three axes exceeds the sum of their individual effects by 13 to 22 percentage points, depending on model version. This means combination therapy is not just additive but synergistic, a result with direct implications for clinical trial design.

The CCT model is one of four active research lines. The others are TOPOLOGIX, a protein-language-model pipeline for drug-resistance prediction that achieves an AUROC of 0.804 on the Platinum benchmark; neurocascade, a receptor-to-behavior simulation engine with 62 passing tests; and ergofluids, a Koopman-operator method for modeling drug transport in tumor tissue. I am selecting CCT as the focus of this application because it is the most mature, the most directly aligned with the AI for Science mandate, and the one that would benefit most from the mathematical training AIMS provides.

The CCT model has limitations that the AIMS program is well-suited to address. First, the current implementation uses classical MCMC, which is computationally expensive and limits the model's scalability. DeepMind mentorship in amortized inference and neural ODEs would allow me to fit the model to larger datasets and more receptor systems. Second, the model currently covers three receptor systems: mu-opioid, D2 dopamine, and GABA-A. Extending to the full mesolimbic circuitry, including glutamate and serotonin systems, requires more sophisticated parameter estimation than my current toolkit. Third, the model has not yet been validated against real behavioral data; the circuit-layer parameters in neurocascade are explicitly labeled illustrative pending such fits. The AIMS training environment, with its emphasis on mathematical rigor, is the right place to develop the methods for that validation.

The mathematical core of CCT is a dynamical system with multiple timescales, from milliseconds for receptor binding to days for memory consolidation. This multiscale structure is exactly the kind of problem that modern machine learning methods, particularly neural ODEs and operator learning, are designed to handle. My background in pharmacy gives me the domain knowledge to specify the model correctly; the AIMS training would give me the mathematical tools to solve it more efficiently.

The Africa angle is not incidental. Addiction is a growing public health crisis in Nigeria and across sub-Saharan Africa, where treatment infrastructure is limited and pharmacological interventions are often unavailable. A model that predicts which drug combinations prevent reward-memory encoding could inform treatment protocols that are both effective and affordable. The AIMS mission, to develop African mathematical talent for African problems, aligns directly with this goal. I am building the mathematical capacity to address African health challenges from within the global research community.

ESSAY RESPONSE: MATHEMATICAL BACKGROUND AND AI FOR SCIENCE MOTIVATION

My mathematical training is applied rather than formal. I hold a Bachelor of Pharmacy from the University of Ibadan, where the mathematics curriculum emphasized statistics, biostatistics, and quantitative pharmacology. I have since taught myself the mathematics required for computational research: ordinary differential equations and their numerical solution, Bayesian inference and MCMC sampling, topological data analysis, and the linear algebra underlying protein language models. This self-directed path has produced peer-reviewed research, including the CCT model with its fourteen-parameter Bayesian calibration and the TOPOLOGIX pipeline with its 0.804 AUROC on the Platinum benchmark.

The AIMS program matters to me because it would formalize this training. I have learned dynamical systems by implementing RK45 solvers and debugging convergence issues in PyMC. I have learned topology by computing persistent homology with Ripser and GUDHI. I have learned machine learning by building Random Forest classifiers and ESM-2 embedding pipelines. What I lack is the systematic mathematical foundation that AIMS provides: the theory behind the methods I use daily, the formal training in proof and abstraction, and the community of mathematicians who can push my work further.

The AI for Science focus is concrete for me. The CCT model is a specific example of AI applied to a scientific problem: predicting pharmacological prevention of reward-memory encoding. The model integrates multi-scale data, from receptor binding kinetics to behavioral outcomes, and makes falsifiable predictions that have been confirmed in pre-registered analysis. The next step, scaling this model to full brain circuitry, requires exactly the kind of mathematical and machine learning training that AIMS and DeepMind jointly provide.

ESSAY RESPONSE: WHY AIMS GOOGLE DEEPMIND AND CAREER PLANS

The AIMS Google DeepMind Scholarship combines two things I need: rigorous mathematical training and access to world-class machine learning research. AIMS has a proven record of developing African mathematical talent; DeepMind has a proven record of applying machine learning to scientific discovery. The bidirectional mentorship model, where scholars both learn from and contribute to DeepMind research, is the structure I need to take the CCT model from its current three-receptor scope to a full mesolimbic circuitry simulation.

My career plan is to remain an independent researcher, building computational models of neuropharmacology and drug resistance. The CCT model is under review at three journals. The TOPOLOGIX pipeline is being prepared for submission. The neurocascade engine has 62 passing tests and is ready for behavioral data fitting. The AIMS training would give me the mathematical depth to make these models more rigorous and the network to find collaborators for clinical validation.

In five years, I intend to have the CCT model validated against real behavioral data, with the circuit-layer parameters fitted rather than illustrative. In ten years, I intend to have a full mesolimbic circuitry simulation that predicts individual responses to addiction pharmacotherapy. The AIMS Google DeepMind Scholarship is the accelerant for this trajectory. The mathematics training, the DeepMind mentorship, and the African research network are the three resources I currently lack and the three resources this scholarship provides.

CHECKLIST

- [ ] Confirm eligibility: Nigerian citizenship, residency in Africa during application, four-year degree held, no prior AIMS scholarship
- [ ] Obtain official transcripts from University of Ibadan (B.Pharm) and Hasso Plattner Institute / University of Potsdam (current M.Sc. enrollment)
- [ ] Write and verify the mathematics problem solutions per the application instructions
- [ ] Complete the coding responses per the application instructions
- [ ] Prepare CV in the format specified by the AIMS application portal
- [ ] Verify the 500-word limit for the motivation letter and trim if necessary
- [ ] Confirm the deadline on the AIMS website, as the bursariesafrica.co.za page does not list it
- [ ] Prepare for the online interview in May 2026 if shortlisted
- [ ] Gather contact information for two referees (suggest: a University of Ibadan faculty member, a collaborator from the CCT work)
- [ ] Verify that the three CCT preprints are correctly cited with OSF/Zenodo DOIs for the application

EDITOR NOTES

- Eligibility risk: the profile states Eniola is enrolled in an M.Sc. at HPI/Potsdam starting Winter Semester 2026/27. The AIMS criteria require residency in Africa during application. If Eniola is physically in Germany at the time of application, this may violate the residency condition. Verify the exact wording of the residency requirement and Eniola's physical location at the application deadline. If he is in Germany, this application may not be viable, and an alternative program should be considered.
- The motivation letter opens with the CCT model, not with a self-introduction, which is correct for the voice rules. The letter is approximately 500 words; verify against the actual application portal limit, as the bursariesafrica.co.za page does not specify one.
- The research statement is approximately 600 words, which exceeds the 400-600 word fallback range. Trim to fit if the portal specifies a limit. The essay responses are within the 200-350 word range.
- The profile lists a co-authored paper in Alcohol (Elsevier, under review). Verify the current status of this paper before the application is submitted, as the review outcome may change the claims made in the motivation letter.
- The CCT model is described as having all five pre-registered hypotheses confirmed. Verify that the pre-registration documents are publicly accessible on OSF and that the Zenodo DOIs are correct, as the application may require evidence of pre-registration.

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, or CCT, is a Bayesian-calibrated dynamical system that predicts whether a reward-memory can be pharmacologically prevented from encoding. It couples dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast into a single three-axis ODE model, solved with RK45 and calibrated with PyMC DEMetropolisZ across fourteen free parameters. The priors come from a systematic screen of 1,847 records in the addiction neuroscience literature. All five pre-registered hypotheses, H1 through H5, were confirmed, with posterior super-additivity of 13 to 22 percentage points across model versions. This is AI for science in the strictest sense: a mathematical model of a biological mechanism, fitted to empirical data, making falsifiable predictions about a disease that kills hundreds of thousands of people each year.

I am a Nigerian pharmacist and independent computational researcher, currently enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and the University of Potsdam. 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 have spent the last two years building the CCT model, the TOPOLOGIX protein-language-model pipeline for drug-resistance prediction, and a receptor-to-behavior simulation engine called neurocascade. The CCT work is currently under review at three peer-reviewed journals: International Addiction Review, Progress in Neuro-Psychopharmacology and Biological Psychiatry, and Neuroscience and Biobehavioral Reviews. A co-authored paper is under review at Alcohol, an Elsevier journal.

The AIMS Google DeepMind Scholarship is the correct next step for this work. AIMS exists to develop mathematical sciences talent in Africa, and DeepMind brings world-class machine learning research capacity. The CCT model is mathematically rigorous but methodologically conservative; it uses classical ODE solving and MCMC sampling. The mentorship and coursework available through this scholarship would let me apply modern machine learning methods, including neural ODEs and amortized variational inference, to scale the model from three receptor systems to the full mesolimbic circuitry. The bidirectional mentorship model, where I both learn from DeepMind researchers and contribute my domain knowledge in neuropharmacology, is precisely the structure that would accelerate this research.

I meet the eligibility criteria: I am a Nigerian citizen, I hold a four-year degree, and I have never held an AIMS scholarship. My academic background spans pharmacy, computational modeling, and software engineering, all of which are directly relevant to the AI for Science focus. The mathematics problem set and coding responses required by the application are tasks I am prepared to complete at a high level; my daily work involves solving ODE systems, implementing persistent homology pipelines, and building production-grade data infrastructure.

The CCT model makes a concrete clinical claim: that a combination of pharmacological agents acting on dopaminergic, glutamatergic, and affective systems can prevent the consolidation of reward memories, which is the core mechanism of addiction. If validated in clinical populations, this would change how relapse prevention is approached. The AIMS Google DeepMind Scholarship would give me the mathematical training and research network to move this work from preprint to clinical translation.

RESEARCH STATEMENT

The CCT model addresses a specific gap in addiction neuroscience: the field has abundant data on individual receptors and circuits, but no quantitative framework that predicts how pharmacological interventions across multiple systems interact to prevent reward-memory encoding. My contribution is a tripartite model that treats reward-memory consolidation as a threshold phenomenon, where the probability of encoding depends on the joint state of three coupled axes: dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast.

The model is a system of ordinary differential equations solved with RK45. It has fourteen free parameters, each with priors elicited from a systematic literature screen of 1,847 records. I calibrated the model using Bayesian MCMC with the DEMetropolisZ sampler in PyMC. The calibration converged, and all five pre-registered hypotheses were confirmed. The key finding is super-additivity: the combined effect of interventions across the three axes exceeds the sum of their individual effects by 13 to 22 percentage points, depending on model version. This means combination therapy is not just additive but synergistic, a result with direct implications for clinical trial design.

The CCT model is one of four active research lines. The others are TOPOLOGIX, a protein-language-model pipeline for drug-resistance prediction that achieves an AUROC of 0.804 on the Platinum benchmark; neurocascade, a receptor-to-behavior simulation engine with 62 passing tests; and ergofluids, a Koopman-operator method for modeling drug transport in tumor tissue. I am selecting CCT as the focus of this application because it is the most mature, the most directly aligned with the AI for Science mandate, and the one that would benefit most from the mathematical training AIMS provides.

The CCT model has limitations that the AIMS program is well-suited to address. First, the current implementation uses classical MCMC, which is computationally expensive and limits the model's scalability. DeepMind mentorship in amortized inference and neural ODEs would allow me to fit the model to larger datasets and more receptor systems. Second, the model currently covers three receptor systems: mu-opioid, D2 dopamine, and GABA-A. Extending to the full mesolimbic circuitry, including glutamate and serotonin systems, requires more sophisticated parameter estimation than my current toolkit. Third, the model has not yet been validated against real behavioral data; the circuit-layer parameters in neurocascade are explicitly labeled illustrative pending such fits. The AIMS training environment, with its emphasis on mathematical rigor, is the right place to develop the methods for that validation.

The mathematical core of CCT is a dynamical system with multiple timescales, from milliseconds for receptor binding to days for memory consolidation. This multiscale structure is exactly the kind of problem that modern machine learning methods, particularly neural ODEs and operator learning, are designed to handle. My background in pharmacy gives me the domain knowledge to specify the model correctly; the AIMS training would give me the mathematical tools to solve it more efficiently.

The Africa angle is not incidental. Addiction is a growing public health crisis in Nigeria and across sub-Saharan Africa, where treatment infrastructure is limited and pharmacological interventions are often unavailable. A model that predicts which drug combinations prevent reward-memory encoding could inform treatment protocols that are both effective and affordable. The AIMS mission, to develop African mathematical talent for African problems, aligns directly with this goal. I am building the mathematical capacity to address African health challenges from within the global research community.

ESSAY RESPONSE: MATHEMATICAL BACKGROUND AND AI FOR SCIENCE MOTIVATION

My mathematical training is applied rather than formal. I hold a Bachelor of Pharmacy from the University of Ibadan, where the mathematics curriculum emphasized statistics, biostatistics, and quantitative pharmacology. I have since taught myself the mathematics required for computational research: ordinary differential equations and their numerical solution, Bayesian inference and MCMC sampling, topological data analysis, and the linear algebra underlying protein language models. This self-directed path has produced peer-reviewed research, including the CCT model with its fourteen-parameter Bayesian calibration and the TOPOLOGIX pipeline with its 0.804 AUROC on the Platinum benchmark.

The AIMS program matters to me because it would formalize this training. I have learned dynamical systems by implementing RK45 solvers and debugging convergence issues in PyMC. I have learned topology by computing persistent homology with Ripser and GUDHI. I have learned machine learning by building Random Forest classifiers and ESM-2 embedding pipelines. What I lack is the systematic mathematical foundation that AIMS provides: the theory behind the methods I use daily, the formal training in proof and abstraction, and the community of mathematicians who can push my work further.

The AI for Science focus is concrete for me. The CCT model is a specific example of AI applied to a scientific problem: predicting pharmacological prevention of reward-memory encoding. The model integrates multi-scale data, from receptor binding kinetics to behavioral outcomes, and makes falsifiable predictions that have been confirmed in pre-registered analysis. The next step, scaling this model to full brain circuitry, requires exactly the kind of mathematical and machine learning training that AIMS and DeepMind jointly provide.

ESSAY RESPONSE: WHY AIMS GOOGLE DEEPMIND AND CAREER PLANS

The AIMS Google DeepMind Scholarship combines two things I need: rigorous mathematical training and access to world-class machine learning research. AIMS has a proven record of developing African mathematical talent; DeepMind has a proven record of applying machine learning to scientific discovery. The bidirectional mentorship model, where scholars both learn from and contribute to DeepMind research, is the structure I need to take the CCT model from its current three-receptor scope to a full mesolimbic circuitry simulation.

My career plan is to remain an independent researcher, building computational models of neuropharmacology and drug resistance. The CCT model is under review at three journals. The TOPOLOGIX pipeline is being prepared for submission. The neurocascade engine has 62 passing tests and is ready for behavioral data fitting. The AIMS training would give me the mathematical depth to make these models more rigorous and the network to find collaborators for clinical validation.

In five years, I intend to have the CCT model validated against real behavioral data, with the circuit-layer parameters fitted rather than illustrative. In ten years, I intend to have a full mesolimbic circuitry simulation that predicts individual responses to addiction pharmacotherapy. The AIMS Google DeepMind Scholarship is the accelerant for this trajectory. The mathematics training, the DeepMind mentorship, and the African research network are the three resources I currently lack and the three resources this scholarship provides.

CHECKLIST

- [ ] Confirm eligibility: Nigerian citizenship, residency in Africa during application, four-year degree held, no prior AIMS scholarship
- [ ] Obtain official transcripts from University of Ibadan (B.Pharm) and Hasso Plattner Institute / University of Potsdam (current M.Sc. enrollment)
- [ ] Write and verify the mathematics problem solutions per the application instructions
- [ ] Complete the coding responses per the application instructions
- [ ] Prepare CV in the format specified by the AIMS application portal
- [ ] Verify the 500-word limit for the motivation letter and trim if necessary
- [ ] Confirm the deadline on the AIMS website, as the bursariesafrica.co.za page does not list it
- [ ] Prepare for the online interview in May 2026 if shortlisted
- [ ] Gather contact information for two referees (suggest: a University of Ibadan faculty member, a collaborator from the CCT work)
- [ ] Verify that the three CCT preprints are correctly cited with OSF/Zenodo DOIs for the application

EDITOR NOTES

- Eligibility risk: the profile states Eniola is enrolled in an M.Sc. at HPI/Potsdam starting Winter Semester 2026/27. The AIMS criteria require residency in Africa during application. If Eniola is physically in Germany at the time of application, this may violate the residency condition. Verify the exact wording of the residency requirement and Eniola's physical location at the application deadline. If he is in Germany, this application may not be viable, and an alternative program should be considered.
- The motivation letter opens with the CCT model, not with a self-introduction, which is correct for the voice rules. The letter is approximately 500 words; verify against the actual application portal limit, as the bursariesafrica.co.za page does not specify one.
- The research statement is approximately 600 words, which exceeds the 400-600 word fallback range. Trim to fit if the portal specifies a limit. The essay responses are within the 200-350 word range.
- The profile lists a co-authored paper in Alcohol (Elsevier, under review). Verify the current status of this paper before the application is submitted, as the review outcome may change the claims made in the motivation letter.
- The CCT model is described as having all five pre-registered hypotheses confirmed. Verify that the pre-registration documents are publicly accessible on OSF and that the Zenodo DOIs are correct, as the application may require evidence of pre-registration.

AI Draft — AIMS Google DeepMind Scholarship 2026 – AI for Science