Draft — EPFL PhD Program in Neuroscience (EDNE)

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, or CCT, is a tripartite pharmacological framework for preventing reward-memory encoding in addiction. It couples three axes, dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast, into a single system of ordinary differential equations solved with RK45. I calibrated the model with Bayesian MCMC using PyMC's DEMetropolisZ sampler across 14 free parameters, with priors elicited from a systematic screen of 1,847 records. All five pre-registered hypotheses, H1 through H5, were confirmed, and the posterior shows super-additivity of 13 to 22 percentage points across model versions. This 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.

The EPFL EDNE PhD program in Neuroscience is the correct environment to develop this research into a full doctoral project. EDNE's explicit focus on computational and systems neuroscience, from molecular mechanisms to circuit-level dynamics, matches the architecture of my work. The CCT model operates at the intersection of receptor pharmacology and dynamical systems theory, and my neurocascade engine extends this by simulating receptor-to-behavior cascades through coupled Wilson-Cowan circuit dynamics. Neurocascade currently has three literature-calibrated receptor systems, mu-opioid, D2 dopamine, and GABA-A, with 62 of 62 tests passing. These are precisely the quantitative, multi-scale methods that EDNE faculty in computational neuroscience and neuroengineering pursue.

My research record demonstrates independent, rigorous execution. I have sole-authored three preprints, each in review at a named journal, and co-authored a paper under review at Alcohol, an Elsevier journal. My work on cardiotoxicity topology tested whether bipartite persistent homology predicts hERG cardiotoxicity from protein-ligand interface geometry. The pre-registered, powered replication found that topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782, settling a comparison the literature had never actually run. I reported this negative result directly. My current TOPOLOGIX project uses ESM-2 protein language model delta-embeddings with Morgan fingerprints and a Random Forest classifier to predict drug-resistance mutations from sequence alone, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark, covering 100 percent of mutations versus roughly 18 percent for structure-limited tools.

Kent Berridge at the University of Michigan has endorsed this research direction. Samuel Gershman at Harvard provided my arXiv endorsement. Nathaniel Daw at Princeton and Marcelo Mattar at NYU are familiar with my computational approach. These are researchers whose work defines the fields of incentive salience and model-based reinforcement learning, and their engagement signals that my framing is credible within the community EDNE draws from.

I am currently enrolled in the M.Sc. in Digital Health at the Hasso Plattner Institute and University of Potsdam, starting Winter Semester 2026/27. My undergraduate degree is a B.Pharm from the University of Ibadan with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9, and I am a PCN-licensed pharmacist. I am applying to EDNE now to establish the trajectory toward doctoral study upon completion of my M.Sc. The program's interdisciplinary scope, from molecular pharmacology to computational neuroscience, is the precise training environment my research program requires.

RESEARCH STATEMENT

My research program is computational neuropharmacology: building mechanistic, quantitative models that connect drug action at receptors to circuit-level dynamics and behavior, with addiction as the primary application domain. The program has two active pillars. The first is the CCT model, a tripartite framework for reward-memory encoding prevention. The second is neurocascade, a receptor-to-behavior simulation engine. Both are Bayesian-calibrated, pre-registered, and built on explicit dynamical-systems foundations.

The CCT model addresses a specific clinical problem: how to pharmacologically prevent the consolidation of reward memories that drive addiction relapse. The model posits that three concurrent processes must be suppressed to block encoding: dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast. I formalized this as a coupled ODE system and calibrated it with Bayesian MCMC using literature-elicited priors from a 1,847-record screen. The model has 14 free parameters. All five pre-registered hypotheses were confirmed. The posterior shows super-additivity of 13 to 22 percentage points, meaning the combined intervention effect exceeds the sum of individual effects. This is a testable, quantitative claim about combination pharmacotherapy. The three sole-authored preprints are under review at International Addiction Review, Progress in Neuro-Psychopharmacology and Biological Psychiatry, and Neuroscience and Biobehavioral Reviews.

Neurocascade extends the CCT framework from molecular targets to circuit dynamics. It couples pharmacokinetic models to receptor binding, then to Wilson-Cowan circuit dynamics, then to behavioral readout layers. I have implemented and calibrated three receptor systems: mu-opioid, D2 dopamine, and GABA-A. The codebase has 62 passing tests. The circuit-layer parameters are explicitly labeled as illustrative pending fits to real behavioral data. This is a deliberate epistemic choice: I do not claim empirical validation I have not performed. The engine is designed to test hypotheses about how receptor-level interventions propagate through circuits to produce behavior, which is exactly the multi-scale question EDNE's computational neuroscience track addresses.

My methodological work includes a rigorous negative result. I tested whether bipartite persistent homology, an opposition-distance metric computed with Ripser and GUDHI, predicts hERG cardiotoxicity from protein-ligand interface geometry. The pre-registered, powered replication found topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. This result matters because the published literature had claimed topological methods were superior without ever running the comparison. I ran it, and I reported the outcome. This is the standard of scientific honesty I bring to EDNE.

The current TOPOLOGIX project applies sequence-based methods to drug-resistance prediction. Using ESM-2 protein language model delta-embeddings, Morgan fingerprints, and a Random Forest classifier, I achieve AUROC 0.804 plus or minus 0.025 on the Platinum benchmark of 553 mutations, and 0.634 on SKEMPI 2.0. This beats structure-based baselines such as mCSM-lig at roughly 0.70 while covering 100 percent of mutations versus approximately 18 percent for structure-limited tools. The lesson from the hERG study, that interface geometry carries less signal than assumed, directly motivated this pivot to sequence representations.

At EDNE, I propose to develop the CCT model into a doctoral project with three aims. First, extend the CCT model to incorporate circuit-level dynamics from neurocascade, creating a unified receptor-to-behavior model of reward-memory encoding. Second, validate the model against behavioral data from rodent models of addiction, using the Bayesian calibration framework I have already established. Third, use the model to generate testable predictions for combination pharmacotherapy, prioritizing interventions with super-additive effects. This program aligns with EDNE faculty working in computational neuroscience, systems neuroscience, and neuroengineering. The program's emphasis on quantitative methods and its location within a leading technical university make it the right environment for this work.

ESSAY: RESEARCH EXPERIENCE AND INDEPENDENCE

My research experience is defined by independent, pre-registered, computationally rigorous projects executed without a formal academic lab. I have built a multi-domain research program spanning addiction neuroscience, protein machine learning, and dynamical-systems methods, and I have done so while working as a clinical pharmacist and product manager.

The CCT model is the centerpiece. I designed the tripartite framework, wrote the ODE solver, performed the Bayesian calibration, and authored all three preprints. The systematic screen of 1,847 records for prior elicitation was my design. The confirmation of all five pre-registered hypotheses was my analysis. The super-additivity finding of 13 to 22 percentage points is my result. This is not a contribution to someone else's project; it is a research program I conceived and executed.

My independence is also demonstrated by my willingness to report negative results. The hERG cardiotoxicity study was a pre-registered, powered replication designed to test whether topological data analysis predicts toxicity from protein-ligand interface geometry. It found that topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I reported this directly. The ergofluids project, which extends Koopman operator methods with a Mori-Zwanzig memory kernel for drug transport in tumor tissue, passed its synthetic-data gates but failed its first real-data gate against digitized published figures. I reported the failure rather than reframing it. This is the discipline EDNE should expect from its doctoral researchers.

My computational skills span the full stack: Python with scipy, numpy, PyMC, and pandas for modeling; Ripser and GUDHI for topological data analysis; NEURON and Brian2 for neural simulation; RDKit and AlphaFold for molecular work; GROMACS and AutoDock for dynamics and docking; and Nextflow, SLURM, and HPC for large-scale computation. I have also built four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech, and social science domains, and I self-host local LLM serving with llama.cpp. This breadth is unusual and directly relevant to EDNE's interdisciplinary scope.

My employment history has supported this research. As a research assistant at CDDDP, I performed NMDA and insulin docking studies. At GHRU-GSAR, I worked on antimicrobial resistance genomics and surveillance pipelines. These positions provided domain exposure, but my independent projects are the evidence of my research potential. I am currently enrolled in the M.Sc. in Digital Health at the Hasso Plattner Institute and University of Potsdam, which will formalize my training in computational methods. My B.Pharm from the University of Ibadan, with a CGPA of 5.1 out of 7.0 and a German equivalent of 1.9, grounds my work in pharmacology. I am a PCN-licensed pharmacist.

CHECKLIST

- [ ] Verify current EDNE application deadline and submission portal on the EPFL EDNE website
- [ ] Confirm eligibility for early-career applicants currently enrolled in an M.Sc. program
- [ ] Obtain three letters of recommendation; confirm Kent Berridge, Samuel Gershman, Nathaniel Daw, and Marcelo Mattar are willing to write
- [ ] Prepare academic transcripts: B.Pharm from University of Ibadan, current M.Sc. enrollment at HPI/Potsdam
- [ ] Prepare CV listing all publications, preprints, and software projects (neurocascade, TOPOLOGIX, CCT)
- [ ] Prepare PDF copies of the three CCT preprints and the Alcohol co-authored paper
- [ ] Prepare a link to the neurocascade GitHub repository with 62 passing tests
- [ ] Prepare a link to the TOPOLOGIX repository with AUROC results on Platinum and SKEMPI 2.0
- [ ] Prepare a statement of purpose specific to EDNE faculty, identifying 2-3 potential supervisors
- [ ] Verify ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) are current and linked
- [ ] Prepare a description of the CCT model's three preprints and their review status at IART, PNPBP, and NBR
- [ ] Confirm the German equivalent grade of 1.9 is correctly calculated and documented
- [ ] Prepare a timeline for M.Sc. completion and proposed EDNE start date
- [ ] Draft responses to any program-specific questions about motivation, fit, and research goals

EDITOR NOTES

- Eligibility risk: The applicant is currently enrolled in an M.Sc. at HPI/Potsdam and has not yet applied for a PhD. EDNE may require a completed M.Sc. or may admit students who are finishing. This must be verified before submission. The letter frames this as establishing a trajectory, which is honest but may not match EDNE's admission timeline.

- The CCT model's three preprints are under review but not yet accepted. The letter states this clearly. If any are rejected before submission, the framing must be updated. Do not present under-review work as published.

- The endorsements from Berridge, Gershman, Daw, and Mattar are listed as collaborators or endorsers in the profile, but the nature and depth of these relationships is not specified. The applicant must confirm these researchers are willing to be named in application materials and, ideally, to write letters. A named endorsement that is not verifiable is a risk.

- The essay on research experience claims independence without a formal lab. This is a strength, but EDNE may expect evidence of ability to work within a lab structure. The applicant should be prepared to address how they will transition from independent work to a collaborative doctoral environment.

- The research statement proposes extending CCT with neurocascade and validating against rodent behavioral data. The applicant has no stated access to a wet lab or animal facility. This aim depends on finding an EDNE faculty member with such resources. The applicant must identify specific faculty and confirm their interest before submission.

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, or CCT, is a tripartite pharmacological framework for preventing reward-memory encoding in addiction. It couples three axes, dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast, into a single system of ordinary differential equations solved with RK45. I calibrated the model with Bayesian MCMC using PyMC's DEMetropolisZ sampler across 14 free parameters, with priors elicited from a systematic screen of 1,847 records. All five pre-registered hypotheses, H1 through H5, were confirmed, and the posterior shows super-additivity of 13 to 22 percentage points across model versions. This 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.

The EPFL EDNE PhD program in Neuroscience is the correct environment to develop this research into a full doctoral project. EDNE's explicit focus on computational and systems neuroscience, from molecular mechanisms to circuit-level dynamics, matches the architecture of my work. The CCT model operates at the intersection of receptor pharmacology and dynamical systems theory, and my neurocascade engine extends this by simulating receptor-to-behavior cascades through coupled Wilson-Cowan circuit dynamics. Neurocascade currently has three literature-calibrated receptor systems, mu-opioid, D2 dopamine, and GABA-A, with 62 of 62 tests passing. These are precisely the quantitative, multi-scale methods that EDNE faculty in computational neuroscience and neuroengineering pursue.

My research record demonstrates independent, rigorous execution. I have sole-authored three preprints, each in review at a named journal, and co-authored a paper under review at Alcohol, an Elsevier journal. My work on cardiotoxicity topology tested whether bipartite persistent homology predicts hERG cardiotoxicity from protein-ligand interface geometry. The pre-registered, powered replication found that topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782, settling a comparison the literature had never actually run. I reported this negative result directly. My current TOPOLOGIX project uses ESM-2 protein language model delta-embeddings with Morgan fingerprints and a Random Forest classifier to predict drug-resistance mutations from sequence alone, achieving AUROC 0.804 plus or minus 0.025 on the Platinum benchmark, covering 100 percent of mutations versus roughly 18 percent for structure-limited tools.

Kent Berridge at the University of Michigan has endorsed this research direction. Samuel Gershman at Harvard provided my arXiv endorsement. Nathaniel Daw at Princeton and Marcelo Mattar at NYU are familiar with my computational approach. These are researchers whose work defines the fields of incentive salience and model-based reinforcement learning, and their engagement signals that my framing is credible within the community EDNE draws from.

I am currently enrolled in the M.Sc. in Digital Health at the Hasso Plattner Institute and University of Potsdam, starting Winter Semester 2026/27. My undergraduate degree is a B.Pharm from the University of Ibadan with a CGPA of 5.1 out of 7.0, a German equivalent of 1.9, and I am a PCN-licensed pharmacist. I am applying to EDNE now to establish the trajectory toward doctoral study upon completion of my M.Sc. The program's interdisciplinary scope, from molecular pharmacology to computational neuroscience, is the precise training environment my research program requires.

RESEARCH STATEMENT

My research program is computational neuropharmacology: building mechanistic, quantitative models that connect drug action at receptors to circuit-level dynamics and behavior, with addiction as the primary application domain. The program has two active pillars. The first is the CCT model, a tripartite framework for reward-memory encoding prevention. The second is neurocascade, a receptor-to-behavior simulation engine. Both are Bayesian-calibrated, pre-registered, and built on explicit dynamical-systems foundations.

The CCT model addresses a specific clinical problem: how to pharmacologically prevent the consolidation of reward memories that drive addiction relapse. The model posits that three concurrent processes must be suppressed to block encoding: dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast. I formalized this as a coupled ODE system and calibrated it with Bayesian MCMC using literature-elicited priors from a 1,847-record screen. The model has 14 free parameters. All five pre-registered hypotheses were confirmed. The posterior shows super-additivity of 13 to 22 percentage points, meaning the combined intervention effect exceeds the sum of individual effects. This is a testable, quantitative claim about combination pharmacotherapy. The three sole-authored preprints are under review at International Addiction Review, Progress in Neuro-Psychopharmacology and Biological Psychiatry, and Neuroscience and Biobehavioral Reviews.

Neurocascade extends the CCT framework from molecular targets to circuit dynamics. It couples pharmacokinetic models to receptor binding, then to Wilson-Cowan circuit dynamics, then to behavioral readout layers. I have implemented and calibrated three receptor systems: mu-opioid, D2 dopamine, and GABA-A. The codebase has 62 passing tests. The circuit-layer parameters are explicitly labeled as illustrative pending fits to real behavioral data. This is a deliberate epistemic choice: I do not claim empirical validation I have not performed. The engine is designed to test hypotheses about how receptor-level interventions propagate through circuits to produce behavior, which is exactly the multi-scale question EDNE's computational neuroscience track addresses.

My methodological work includes a rigorous negative result. I tested whether bipartite persistent homology, an opposition-distance metric computed with Ripser and GUDHI, predicts hERG cardiotoxicity from protein-ligand interface geometry. The pre-registered, powered replication found topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. This result matters because the published literature had claimed topological methods were superior without ever running the comparison. I ran it, and I reported the outcome. This is the standard of scientific honesty I bring to EDNE.

The current TOPOLOGIX project applies sequence-based methods to drug-resistance prediction. Using ESM-2 protein language model delta-embeddings, Morgan fingerprints, and a Random Forest classifier, I achieve AUROC 0.804 plus or minus 0.025 on the Platinum benchmark of 553 mutations, and 0.634 on SKEMPI 2.0. This beats structure-based baselines such as mCSM-lig at roughly 0.70 while covering 100 percent of mutations versus approximately 18 percent for structure-limited tools. The lesson from the hERG study, that interface geometry carries less signal than assumed, directly motivated this pivot to sequence representations.

At EDNE, I propose to develop the CCT model into a doctoral project with three aims. First, extend the CCT model to incorporate circuit-level dynamics from neurocascade, creating a unified receptor-to-behavior model of reward-memory encoding. Second, validate the model against behavioral data from rodent models of addiction, using the Bayesian calibration framework I have already established. Third, use the model to generate testable predictions for combination pharmacotherapy, prioritizing interventions with super-additive effects. This program aligns with EDNE faculty working in computational neuroscience, systems neuroscience, and neuroengineering. The program's emphasis on quantitative methods and its location within a leading technical university make it the right environment for this work.

ESSAY: RESEARCH EXPERIENCE AND INDEPENDENCE

My research experience is defined by independent, pre-registered, computationally rigorous projects executed without a formal academic lab. I have built a multi-domain research program spanning addiction neuroscience, protein machine learning, and dynamical-systems methods, and I have done so while working as a clinical pharmacist and product manager.

The CCT model is the centerpiece. I designed the tripartite framework, wrote the ODE solver, performed the Bayesian calibration, and authored all three preprints. The systematic screen of 1,847 records for prior elicitation was my design. The confirmation of all five pre-registered hypotheses was my analysis. The super-additivity finding of 13 to 22 percentage points is my result. This is not a contribution to someone else's project; it is a research program I conceived and executed.

My independence is also demonstrated by my willingness to report negative results. The hERG cardiotoxicity study was a pre-registered, powered replication designed to test whether topological data analysis predicts toxicity from protein-ligand interface geometry. It found that topological features do not beat a plain descriptor baseline, AUROC 0.8426 versus 0.8782. I reported this directly. The ergofluids project, which extends Koopman operator methods with a Mori-Zwanzig memory kernel for drug transport in tumor tissue, passed its synthetic-data gates but failed its first real-data gate against digitized published figures. I reported the failure rather than reframing it. This is the discipline EDNE should expect from its doctoral researchers.

My computational skills span the full stack: Python with scipy, numpy, PyMC, and pandas for modeling; Ripser and GUDHI for topological data analysis; NEURON and Brian2 for neural simulation; RDKit and AlphaFold for molecular work; GROMACS and AutoDock for dynamics and docking; and Nextflow, SLURM, and HPC for large-scale computation. I have also built four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech, and social science domains, and I self-host local LLM serving with llama.cpp. This breadth is unusual and directly relevant to EDNE's interdisciplinary scope.

My employment history has supported this research. As a research assistant at CDDDP, I performed NMDA and insulin docking studies. At GHRU-GSAR, I worked on antimicrobial resistance genomics and surveillance pipelines. These positions provided domain exposure, but my independent projects are the evidence of my research potential. I am currently enrolled in the M.Sc. in Digital Health at the Hasso Plattner Institute and University of Potsdam, which will formalize my training in computational methods. My B.Pharm from the University of Ibadan, with a CGPA of 5.1 out of 7.0 and a German equivalent of 1.9, grounds my work in pharmacology. I am a PCN-licensed pharmacist.

CHECKLIST

- [ ] Verify current EDNE application deadline and submission portal on the EPFL EDNE website
- [ ] Confirm eligibility for early-career applicants currently enrolled in an M.Sc. program
- [ ] Obtain three letters of recommendation; confirm Kent Berridge, Samuel Gershman, Nathaniel Daw, and Marcelo Mattar are willing to write
- [ ] Prepare academic transcripts: B.Pharm from University of Ibadan, current M.Sc. enrollment at HPI/Potsdam
- [ ] Prepare CV listing all publications, preprints, and software projects (neurocascade, TOPOLOGIX, CCT)
- [ ] Prepare PDF copies of the three CCT preprints and the Alcohol co-authored paper
- [ ] Prepare a link to the neurocascade GitHub repository with 62 passing tests
- [ ] Prepare a link to the TOPOLOGIX repository with AUROC results on Platinum and SKEMPI 2.0
- [ ] Prepare a statement of purpose specific to EDNE faculty, identifying 2-3 potential supervisors
- [ ] Verify ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) are current and linked
- [ ] Prepare a description of the CCT model's three preprints and their review status at IART, PNPBP, and NBR
- [ ] Confirm the German equivalent grade of 1.9 is correctly calculated and documented
- [ ] Prepare a timeline for M.Sc. completion and proposed EDNE start date
- [ ] Draft responses to any program-specific questions about motivation, fit, and research goals

EDITOR NOTES

- Eligibility risk: The applicant is currently enrolled in an M.Sc. at HPI/Potsdam and has not yet applied for a PhD. EDNE may require a completed M.Sc. or may admit students who are finishing. This must be verified before submission. The letter frames this as establishing a trajectory, which is honest but may not match EDNE's admission timeline.

- The CCT model's three preprints are under review but not yet accepted. The letter states this clearly. If any are rejected before submission, the framing must be updated. Do not present under-review work as published.

- The endorsements from Berridge, Gershman, Daw, and Mattar are listed as collaborators or endorsers in the profile, but the nature and depth of these relationships is not specified. The applicant must confirm these researchers are willing to be named in application materials and, ideally, to write letters. A named endorsement that is not verifiable is a risk.

- The essay on research experience claims independence without a formal lab. This is a strength, but EDNE may expect evidence of ability to work within a lab structure. The applicant should be prepared to address how they will transition from independent work to a collaborative doctoral environment.

- The research statement proposes extending CCT with neurocascade and validating against rodent behavioral data. The applicant has no stated access to a wet lab or animal facility. This aim depends on finding an EDNE faculty member with such resources. The applicant must identify specific faculty and confirm their interest before submission.

AI Draft — EPFL PhD Program in Neuroscience (EDNE)