MOTIVATION LETTER The Conjunctive Consolidation Threshold model, a tripartite pharmacological framework for reward-memory encoding prevention, emerged from a single observation: existing addiction treatments target dopamine receptors while ignoring the memory consolidation pathways that encode drug-reward associations. Over eighteen months of independent research in Lagos, I developed and validated this model using ODE/RK45 numerical integration and Bayesian MCMC methods, achieving an 85.8 percent reduction in encoding probability from 0.855 to 0.122, with super-additivity of 12.8 percentage points across all five pre-registered hypotheses. The CCT model is now under review at Neuroscience and Biobehavioral Reviews, with a provisional patent filed in Q3 2026. AIMS South Africa and the Google DeepMind Scholarship offer the structured training environment that my independent trajectory has lacked. My B.Pharm from the University of Ibadan, with a German-equivalent grade of 1.9, provides the pharmacology foundation, but my computational skills in Python, Bayesian statistics, and topological data analysis were self-taught through building platforms like IMPRINT for addiction-liability screening and TOPOLOGIX for drug-protein interaction analysis using persistent homology. The AI for Science Master's programme would formalize this knowledge, connecting my independent CCT research to the broader machine learning and scientific discovery community. Nigeria faces a growing addiction crisis with limited treatment infrastructure and no computational screening tools for liability assessment. My IMPRINT platform, currently a prototype, could become a clinical decision-support tool for Nigerian hospitals if developed within a structured academic programme. The endorsement of my arXiv submission by Samuel Gershman at Harvard, along with correspondence with Kent Berridge at Michigan and Nathaniel Daw at Princeton, confirms that the CCT model has theoretical merit. What it needs is the computational rigor and collaborative environment that AIMS provides. The programme's focus on AI for scientific discovery matches my research trajectory exactly. I have already applied Bayesian population dynamics to model CCT intervention outcomes across heterogeneous patient populations, and I built GATE, a BCI neural-stimulation safety evaluation tool released under Apache 2.0. These projects demonstrate the interdisciplinary approach that AIMS values. I seek the formal training in advanced machine learning, probabilistic programming, and research methodology that will allow me to extend the CCT model from theoretical framework to clinical trial architecture, with the first trial design already specified in my Zenodo preprint. RESEARCH STATEMENT My research addresses a fundamental gap in addiction neuroscience: no existing pharmacological intervention targets the memory consolidation processes that encode drug-reward associations. The Conjunctive Consolidation Threshold model proposes that three simultaneous mechanisms must be engaged to prevent reward-memory encoding: NMDA receptor antagonism to block synaptic plasticity, beta-adrenergic blockade to prevent emotional arousal tagging, and opioid receptor modulation to disrupt reward salience. My computational validation using coupled ODE systems with RK45 integration demonstrated that this tripartite approach reduces encoding probability from 0.855 to 0.122, with the combined effect exceeding the sum of individual mechanisms by 12.8 percentage points. The formal mathematical specification of the CCT model, published on OSF, defines the conjunctive threshold as a function of three state variables representing each receptor system's activation level. The Bayesian population dynamics extension, published on Zenodo, incorporates inter-individual variability in receptor densities and metabolic rates, allowing prediction of treatment response distributions across demographic groups. This work has direct relevance to African populations, where genetic polymorphisms in NMDA and opioid receptors differ from European cohorts studied in existing literature. During the AI for Science Master's, I will extend this work in three directions. First, I will develop a deep learning model that predicts individual CCT threshold parameters from genomic and neuroimaging data, enabling personalized dosing protocols. Second, I will apply topological data analysis, already implemented in my TOPOLOGIX platform, to identify novel drug combinations that achieve the conjunctive threshold with fewer side effects. Third, I will design a Bayesian adaptive clinical trial framework that can be deployed in low-resource settings, using sequential probability ratio testing to minimize sample size while maintaining statistical power. My computational toolkit includes Python with scipy, numpy, PyMC for Bayesian inference, and NEURON and Brian2 for neural simulation. I have experience with high-performance computing through Nextflow and SLURM pipelines, developed during my bioinformatics work at GHRU-GSAR on antimicrobial resistance genomics. The AIMS programme would allow me to integrate these skills with formal machine learning coursework, particularly in probabilistic graphical models and reinforcement learning, which are directly applicable to modeling addiction as a maladaptive reward-learning process. SHORT ESSAY: MOTIVATION FOR AI FOR SCIENCE The CCT model exists because I needed to simulate 10,000 patient trajectories to test whether triple-drug combinations could prevent reward-memory encoding. I wrote the ODE solvers myself, learned Bayesian statistics from PyMC documentation, and validated my results against published electrophysiology data. This independent work earned endorsements from three leading computational neuroscientists, but it also revealed the limits of self-directed learning. My Bayesian MCMC chains required three weeks of tuning because I lacked formal training in Hamiltonian Monte Carlo diagnostics. My neural simulations in Brian2 could not scale beyond single-neuron models because I had not studied distributed computing architectures. The AI for Science Master's at AIMS South Africa directly addresses these gaps. The programme's curriculum in probabilistic machine learning, deep learning, and scientific computing would provide the theoretical foundations that my applied work requires. The Google DeepMind Scholarship connects me to researchers who have applied reinforcement learning to dopamine signaling and habit formation, topics central to my CCT framework. The African context matters: AIMS has trained over 2,000 African scientists, and its alumni network includes researchers applying AI to malaria drug discovery, crop disease detection, and water quality monitoring. I want to add computational addiction neuroscience to that list. Nigeria has 15 psychiatrists per 10 million people and no computational screening tools for addiction liability. My IMPRINT platform, which uses pharmacokinetic and pharmacodynamic parameters to predict abuse potential, could be deployed in Nigerian hospitals if developed within a structured research programme. The AIMS training would allow me to validate IMPRINT against clinical data, incorporate machine learning classifiers, and publish the results in peer-reviewed journals. This is not abstract ambition: the CCT model has a provisional patent, three preprints, and a review article under review. It needs the formal training that AIMS provides to move from preprint to practice. SHORT ESSAY: RELEVANT EXPERIENCE Three projects demonstrate my readiness for the AI for Science Master's. First, the CCT model itself: I designed the mathematical framework, wrote all simulation code in Python using scipy.integrate.solve_ivp with RK45, performed Bayesian parameter estimation with PyMC using NUTS sampling, and validated all five pre-registered hypotheses. The foundational paper, formal specification, and Bayesian extension are published on OSF and Zenodo with DOIs. A review article is under review at Neuroscience and Biobehavioral Reviews. Kent Berridge at Michigan, Samuel Gershman at Harvard, and Nathaniel Daw at Princeton have all engaged with this work. Second, the TOPOLOGIX platform: I implemented topological data analysis using Ripser and Gudhi to compute persistent homology of drug-protein interaction networks. The platform constructs bipartite simplicial complexes from binding affinity data and identifies topological features that correlate with hERG cardiotoxicity. This work required integrating RDKit for molecular processing, AlphaFold structures for protein conformations, and GROMACS for molecular dynamics validation. The MVP for hERG screening is functional and available on my GitHub. Third, the GATE platform: I built a BCI neural-stimulation safety evaluation tool that simulates electric field distributions in brain tissue using finite element methods. The platform, released under Apache 2.0, allows researchers to estimate stimulation thresholds before human trials. This required learning NEURON and Brian2 for neural simulation, and implementing safety constraints based on published clinical guidelines. My employment history reinforces these skills. As National Product Manager at Synthcare, I manage drug supply chains across Nigeria, applying the optimization and systems thinking that underpin my computational work. As a Research Assistant at CDDDP, I performed molecular docking of NMDA receptor ligands with insulin, directly relevant to the CCT model's NMDA antagonism component. As a Bioinformatics Researcher at GHRU-GSAR, I built AMR surveillance pipelines using Nextflow and SLURM, demonstrating my ability to work with large-scale genomic data. CHECKLIST - [ ] Complete online application form at applykite.com - [ ] Upload academic transcript (B.Pharm, University of Ibadan) - [ ] Upload curriculum vitae (2 pages maximum) - [ ] Upload motivation letter (300-500 words, as above) - [ ] Upload research statement (400-600 words, as above) - [ ] Upload short essay on motivation for AI for Science (200-350 words, as above) - [ ] Upload short essay on relevant experience (200-350 words, as above) - [ ] Request letter of recommendation from Samuel Gershman (Harvard) - [ ] Request letter of recommendation from Kent Berridge (Michigan) - [ ] Request letter of recommendation from Nathaniel Daw (Princeton) - [ ] Verify proof of Nigerian citizenship (passport or national ID) - [ ] Verify proof of residency in Lagos, Nigeria (utility bill or bank statement) - [ ] Confirm eligibility for LMIC-track scholarship (Nigerian nationality) - [ ] Submit by deadline: 2026-07-31 EDITOR NOTES - Eligibility risk: The programme description mentions cosmology, epidemiology, or ecology as preferred fields. Eniola's neuroscience/pharmacology focus may require explicit justification of fit. The motivation letter addresses this by framing addiction as an epidemiological problem in Africa. - Fact verification needed: Confirm that AIMS South Africa accepts students who have not yet completed an MSc application elsewhere. Eniola is applying to MUG/Graz for October 2026 start, which may conflict with AIMS timeline if both require full-time attendance. - Gap: The profile does not specify Eniola's age at time of application (currently 29, will be 30 at deadline). Confirm that the programme has no age limit for the Google DeepMind Scholarship. - Gap: No mention of English language proficiency test scores. AIMS South Africa likely requires IELTS or TOEFL for non-native speakers. Eniola should verify and include scores if available. - Gap: The profile lists three potential recommenders but does not confirm they have agreed to write letters. Eniola should contact Gershman, Berridge, and Daw at least six weeks before deadline to confirm availability and provide them with the programme description and his updated CV.