MOTIVATION LETTER The CCT model, a tripartite pharmacological framework for reward-memory encoding prevention in addiction, was built with open-source code, pre-registered hypotheses, and Bayesian MCMC calibration using literature-elicited priors from an 1,847-record screen. All five pre-registered hypotheses were confirmed, with posterior super-additivity of 13-22 percentage points across model versions. The model is preprinted on OSF and Zenodo, and a co-authored paper is under review at Alcohol. This project, like my TOPOLOGIX drug-resistance mutation predictor and my neurocascade brain-circuit simulation engine, exists because I work as an independent researcher with no institutional lab or grant overhead. AxonDAO's DeSci mission directly supports this mode of work: decentralized, community-driven health research that prioritizes open sharing over proprietary gatekeeping. AxonOS compute credits would accelerate two specific bottlenecks. First, the neurocascade engine couples pharmacokinetics to receptor-binding to Wilson-Cowan circuit dynamics across three literature-calibrated systems (mu-opioid, D2 dopamine, GABA-A). Bayesian calibration of the circuit-layer parameters currently runs on my personal hardware; scaling to real behavioral-data fits requires the kind of parallel MCMC sampling that AxonOS can provide. Second, TOPOLOGIX uses ESM-2 protein-language-model delta-embeddings and Morgan/ECFP fingerprints with a Random Forest classifier. It achieves AUROC 0.804 on the Platinum benchmark (553 mutations) and 0.634 on SKEMPI 2.0, covering 100% of mutations versus roughly 18% for structure-limited tools. Scaling to the full SKEMPI 2.0 dataset and beyond requires compute that exceeds my current capacity. My Nigerian background and independent-researcher status are the reason the work is structured as it is: fully open, reproducible, and designed to function without institutional infrastructure. AxonDAO's rolling, lightweight review process and emphasis on what an applicant will build and share openly align with how I have operated since 2024. I am enrolled in the M.Sc. Digital Health programme at Hasso Plattner Institute starting Winter Semester 2026/27, which will formalize my training in digital health methods, but the research itself has been and will remain independent and openly shared. RESEARCH STATEMENT My research spans three domains that share a common methodological core: dynamical-systems modeling, Bayesian calibration, and open reproducible pipelines. The CCT model addresses addiction at the circuit-pharmacology level. It is a three-axis ordinary differential equation system (dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, affective contrast) solved with RK45 and calibrated with PyMC's DEMetropolisZ sampler across 14 free parameters. All five pre-registered hypotheses were confirmed. The model predicts that conjunctive consolidation threshold crossing can be prevented by targeting any two of the three axes simultaneously, with super-additive effect. A co-authored paper is under review at Alcohol. The TOPOLOGIX project predicts drug-resistance mutations from sequence alone. Structure-based tools like mCSM-lig cover roughly 18% of mutations because they require a resolved protein-ligand complex. TOPOLOGIX uses ESM-2 embeddings and Morgan fingerprints with a Random Forest classifier, achieving AUROC 0.804 on the Platinum benchmark and 0.634 on SKEMPI 2.0. It covers 100% of mutations. This project grew directly from my earlier work on interface topology for hERG cardiotoxicity prediction, where I discovered that bipartite persistent homology does not beat a plain descriptor baseline (AUROC 0.8426 versus 0.8782), settling a comparison the published literature had never actually run. The neurocascade engine is the most computationally demanding of the three. It couples pharmacokinetics to receptor-binding to Wilson-Cowan circuit dynamics to behavioral-readout ODE layers. Three receptor/circuit systems are calibrated from literature data: mu-opioid, D2 dopamine, and GABA-A. The engine passes 62 of 62 tests, but the circuit-layer parameters are explicitly labeled illustrative pending real behavioral-data fits. AxonOS compute credits would allow me to run the Bayesian calibration on real datasets rather than synthetic or literature-derived ones. All three projects are pre-registered, preprinted, and built with reproducible pipelines. Code is on GitHub (github.com/AmunRaPtah). Data and analysis scripts are on OSF and Zenodo. Independent research without institutional support requires that every step be transparent and reproducible, because there is no lab to absorb errors or retractions. PROJECT PROPOSAL I request AxonDAO compute credits and mentorship to complete two milestones over a six-month period. Milestone one: Calibrate the neurocascade engine to real behavioral data from published rodent self-administration studies. The current engine passes 62 of 62 synthetic tests but has not been fit to actual behavioral readouts. I will extract time-series data from three published studies (one per receptor system), run Bayesian MCMC calibration with PyMC on AxonOS, and compare model fit against a null model. Success criterion: posterior predictive checks show the model captures at least 80% of the variance in the behavioral readout for at least two of the three systems. All code, data extraction scripts, and calibrated parameters will be shared under an MIT license on GitHub and archived on Zenodo. Milestone two: Scale TOPOLOGIX to the full SKEMPI 2.0 dataset and add a second classifier (gradient boosting) for comparison. Current performance is AUROC 0.634 on SKEMPI 2.0. I will run hyperparameter sweeps across both classifiers, test feature ablation (ESM-2 embeddings only, Morgan fingerprints only, combined), and report precision-recall curves alongside AUROC. Success criterion: achieve AUROC above 0.70 on SKEMPI 2.0 with the best-performing configuration. All model weights, training scripts, and evaluation results will be shared under an MIT license. Mentorship from AxonDAO's network of AI and biology researchers would be valuable for both milestones. For neurocascade, I need guidance on selecting appropriate behavioral datasets and defining likelihood functions that match the noise structure of rodent self-administration data. For TOPOLOGIX, I want feedback on whether protein language model embeddings are the right representation for this task or whether graph-based representations of the binding interface would be more informative. BUDGET AND RESOURCES No USD amount is specified for AxonDAO grants. The primary resource requested is compute credits on AxonOS. Estimated compute requirements: 500 GPU-hours for neurocascade calibration (MCMC sampling across three systems with 100,000 iterations each) and 200 GPU-hours for TOPOLOGIX scaling (hyperparameter sweeps across two classifiers and three feature sets). Total: 700 GPU-hours. I have my own storage and data management infrastructure (self-hosted Supabase/Postgres, DuckDB-based pipelines, automated backup and disaster recovery on a Linux VPS with systemd and Caddy TLS). No additional hardware or cloud storage is needed. CHECKLIST - [ ] Complete the AxonDAO grant application form at https://www.axondao.io/grants - [ ] Upload this motivation letter as a PDF - [ ] Upload this research statement as a PDF - [ ] Upload this project proposal as a PDF - [ ] Upload this budget and resources statement as a PDF - [ ] Provide links to GitHub profile (github.com/AmunRaPtah), ORCID (0009-0001-9272-6735), and personal site (zyco.org) - [ ] Provide links to preprints on OSF and Zenodo for the CCT model - [ ] Provide links to pre-registrations for the CCT model, hERG topology study, and ergofluids project - [ ] Confirm that all code and data will be shared under an open license (MIT for code, CC-BY for data) - [ ] Confirm that the project will be built in an AxonOS-compatible environment for reproducibility EDITOR NOTES - Eligibility risk: AxonDAO grants are rolling and do not specify geographic restrictions, but the applicant should confirm that Nigerian citizenship and independent researcher status are explicitly permitted. The application form may ask for institutional affiliation; if so, list "Independent researcher" and note the upcoming HPI enrollment. - Fact verification: The applicant states that a co-authored paper is under review at Alcohol. Confirm the current status of this submission before submitting the application. If the paper has been accepted or rejected, update the language accordingly. - Gap: The applicant's employment history includes a role as National Product Manager at Synthcare starting March 2026. The application should clarify whether this role is full-time and how it will affect the timeline for the proposed milestones. If the role is part-time or flexible, state that explicitly. - Gap: The applicant's M.Sc. Digital Health programme at HPI starts in Winter Semester 2026/27. The application should clarify whether the applicant will have time for independent research during the programme, or whether the proposed milestones will be completed before enrollment begins. - Gap: The applicant's ergofluids project is mentioned in the profile but not in the application materials. If the applicant wants to include it, it should be added to the research statement or project proposal. If not, remove it from the profile summary to avoid confusion.