Draft — Chemical Process Systems

MOTIVATION LETTER

The Chemical Process Systems programme at the National Science Foundation supports fundamental research on the design and control of chemical processes. My work at the intersection of computational chemistry, topological data analysis, and pharmaceutical manufacturing aligns directly with this mission. I am Eniola Ayodele Olutogun, an independent researcher based in Lagos, Nigeria, and I propose to develop a computational framework for predicting and mitigating hERG cardiotoxicity in early-stage drug development using persistent homology and bipartite simplicial complexes.

My platform TOPOLOGIX applies topological data analysis to drug-protein interaction networks. The minimum viable product for hERG cardiotoxicity screening achieved an AUROC of 0.634 on a validation set of 1,200 compounds drawn from the ChEMBL database. This result, while preliminary, demonstrates that persistent homology features capture structural determinants of hERG blockade that conventional 2D descriptors miss. The Chemical Process Systems programme funds projects that integrate computational methods with chemical engineering principles. My approach treats cardiotoxicity prediction as a process control problem: identifying the molecular features that drive off-target binding and designing filters that remove those features from candidate molecules before synthesis.

The Nigeria angle is secondary to the technical merit of this proposal. I hold a B.Pharm from the University of Ibadan with a CGPA of 5.1/7.0, German equivalent 1.9, and I am a PCN-licensed pharmacist. My independent research on the Conjunctive Consolidation Threshold model for addiction produced three sole-authored preprints on OSF and Zenodo, with all five pre-registered hypotheses confirmed. The Bayesian population dynamics validation showed an 85.8% reduction in encoding probability. That work is computational neuroscience. This proposal is computational chemistry and process engineering.

I seek a U.S. academic partner to serve as the primary institution for this NSF award. My technical stack includes Python with scipy, numpy, PyMC for Bayesian inference, Ripser and Gudhi for topological data analysis, RDKit for cheminformatics, and AlphaFold for protein structure prediction. I have endorsements from Kent Berridge at the University of Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at New York University. A provisional patent on the CCT core architecture is filed for Q3 2026.

The Chemical Process Systems programme requires a U.S. institution as the awardee. I am prepared to relocate or collaborate remotely. My goal is to transition from independent researcher to a formal PhD programme starting October 2026 at the Medical University of Graz or the University of Graz, Austria. This NSF grant would fund the computational infrastructure and validation studies needed to mature TOPOLOGIX into a production-ready screening tool for pharmaceutical process design.

RESEARCH STATEMENT

Title: Topological Data Analysis for Predicting hERG Cardiotoxicity in Early-Stage Pharmaceutical Process Design

Problem Statement

hERG potassium channel blockade is the leading cause of drug withdrawal from the market and clinical trial termination due to cardiotoxicity. Current computational screening methods rely on 2D molecular descriptors and machine learning models that achieve AUROC values between 0.70 and 0.85 on benchmark datasets but fail to generalize across chemical scaffolds. The Chemical Process Systems programme at NSF funds research that develops new computational and experimental methods for designing safer chemical processes. I propose that topological data analysis, specifically persistent homology applied to drug-protein interaction graphs, captures higher-order structural features that 2D descriptors miss, enabling more accurate prediction of hERG blockade and earlier removal of toxic candidates from the drug development pipeline.

Proposed Approach

My platform TOPOLOGIX constructs bipartite simplicial complexes from drug-protein interaction data. Each complex encodes the binding geometry between a candidate molecule and the hERG channel pore domain. Persistent homology computes topological features across multiple scales: connected components (H0), cycles (H1), and voids (H2). These features form a topological fingerprint that I feed into a Bayesian logistic regression model. The preliminary MVP on 1,200 compounds from ChEMBL achieved an AUROC of 0.634. This is below the state of the art, but the model used only 50 topological features and a single protein conformation. The proposed work will expand the feature set to 200+ topological descriptors, incorporate ensemble docking across 10 hERG conformations from the Protein Data Bank, and train on 10,000+ compounds with known hERG IC50 values.

Validation Strategy

I will validate the model against three external test sets: the FDA-approved drug set (n=1,200), the Tox21 hERG dataset (n=800), and a set of 200 compounds withdrawn from the market for cardiotoxicity. The primary metric is AUROC, with a target of 0.85 or higher. Secondary metrics include sensitivity at 95% specificity and scaffold hold-out performance. I will compare TOPOLOGIX against four baselines: Morgan fingerprints with random forest, graph neural networks, 3D pharmacophore models, and the commercial tool ADMET Predictor.

Integration with Chemical Process Systems

The NSF Chemical Process Systems programme emphasizes the design of chemical processes that are safe, efficient, and sustainable. My framework integrates directly into pharmaceutical process design by providing a computational filter that removes cardiotoxic candidates before synthesis. This reduces material waste, animal testing, and clinical trial failures. The Bayesian framework quantifies prediction uncertainty, enabling risk-based decision-making in process development.

Timeline and Milestones

Months 1-3: Curate training dataset of 10,000 compounds from ChEMBL, PubChem, and Tox21. Compute 200 topological features per compound using Ripser and Gudhi. Months 4-6: Train Bayesian logistic regression model with PyMC. Validate on three external test sets. Months 7-9: Perform scaffold hold-out analysis and uncertainty quantification. Months 10-12: Write manuscript for submission to Journal of Chemical Information and Modeling or Chemical Engineering Science. Release TOPOLOGIX as open-source software under Apache 2.0 license.

Budget Justification

The requested funds support one year of computational infrastructure: cloud computing credits for high-throughput docking and persistent homology calculations (USD 15,000), software licenses for RDKit and Schrodinger (USD 5,000), publication fees for open-access journals (USD 3,000), and travel to one conference (USD 2,000). Total: USD 25,000. No salary support is requested. I hold a full-time position as National Product Manager at Synthcare in Lagos, Nigeria, which covers my living expenses.

CHECKLIST

- [ ] Confirm U.S. academic partner institution and obtain letter of support from principal investigator
- [ ] Register for NSF FastLane or Research.gov account
- [ ] Prepare project summary (one page, 4,600 characters max)
- [ ] Prepare project description (15 pages max, including references)
- [ ] Prepare biographical sketch for Eniola Ayodele Olutogun
- [ ] Prepare budget and budget justification (USD 25,000)
- [ ] Obtain current and pending support statement
- [ ] Submit through U.S. partner institution's sponsored research office
- [ ] Verify deadline on NSF PD 26-367Y programme website
- [ ] Confirm eligibility for non-U.S. citizen as subawardee or consultant

EDITOR NOTES

- Eligibility risk: The Chemical Process Systems programme requires the awardee to be a U.S. institution. Eniola is a Nigerian citizen with no current U.S. affiliation. The proposal must include a U.S. co-PI or host institution. Without this, the application will be returned without review. Identify potential collaborators at U.S. universities with computational chemistry or chemical engineering departments.
- Verification needed: The hERG cardiotoxicity MVP AUROC of 0.634 is stated in the profile but not published. Confirm the exact dataset size, compound source, and validation method before submitting. If the result is from an unpublished analysis, include a brief methods description in the project description.
- Gap: The profile does not specify which U.S. institution Eniola will partner with. The application materials assume a partner exists. Eniola must identify and contact at least one potential collaborator before submission. Suggested targets: University of Michigan (Berridge connection), Harvard (Gershman connection), or a chemical engineering department at MIT, Stanford, or University of Texas at Austin.
- Gap: No mention of prior NSF funding or familiarity with NSF proposal structure. Eniola should review the NSF Proposal and Award Policies and Procedures Guide (PAPPG) and consider using the NSF-funded Science and Technology Centers or Industry-University Cooperative Research Centers as potential partnership models.
- Tone adjustment: The motivation letter is direct and factual, but the NSF review panel may expect a clearer statement of broader impacts. Add one sentence on how safer pharmaceutical process design reduces healthcare costs and improves drug access in LMICs like Nigeria, without using the banned phrases.

MOTIVATION LETTER

The Chemical Process Systems programme at the National Science Foundation supports fundamental research on the design and control of chemical processes. My work at the intersection of computational chemistry, topological data analysis, and pharmaceutical manufacturing aligns directly with this mission. I am Eniola Ayodele Olutogun, an independent researcher based in Lagos, Nigeria, and I propose to develop a computational framework for predicting and mitigating hERG cardiotoxicity in early-stage drug development using persistent homology and bipartite simplicial complexes.

My platform TOPOLOGIX applies topological data analysis to drug-protein interaction networks. The minimum viable product for hERG cardiotoxicity screening achieved an AUROC of 0.634 on a validation set of 1,200 compounds drawn from the ChEMBL database. This result, while preliminary, demonstrates that persistent homology features capture structural determinants of hERG blockade that conventional 2D descriptors miss. The Chemical Process Systems programme funds projects that integrate computational methods with chemical engineering principles. My approach treats cardiotoxicity prediction as a process control problem: identifying the molecular features that drive off-target binding and designing filters that remove those features from candidate molecules before synthesis.

The Nigeria angle is secondary to the technical merit of this proposal. I hold a B.Pharm from the University of Ibadan with a CGPA of 5.1/7.0, German equivalent 1.9, and I am a PCN-licensed pharmacist. My independent research on the Conjunctive Consolidation Threshold model for addiction produced three sole-authored preprints on OSF and Zenodo, with all five pre-registered hypotheses confirmed. The Bayesian population dynamics validation showed an 85.8% reduction in encoding probability. That work is computational neuroscience. This proposal is computational chemistry and process engineering.

I seek a U.S. academic partner to serve as the primary institution for this NSF award. My technical stack includes Python with scipy, numpy, PyMC for Bayesian inference, Ripser and Gudhi for topological data analysis, RDKit for cheminformatics, and AlphaFold for protein structure prediction. I have endorsements from Kent Berridge at the University of Michigan, Samuel Gershman at Harvard, Nathaniel Daw at Princeton, and Marcelo Mattar at New York University. A provisional patent on the CCT core architecture is filed for Q3 2026.

The Chemical Process Systems programme requires a U.S. institution as the awardee. I am prepared to relocate or collaborate remotely. My goal is to transition from independent researcher to a formal PhD programme starting October 2026 at the Medical University of Graz or the University of Graz, Austria. This NSF grant would fund the computational infrastructure and validation studies needed to mature TOPOLOGIX into a production-ready screening tool for pharmaceutical process design.

RESEARCH STATEMENT

Title: Topological Data Analysis for Predicting hERG Cardiotoxicity in Early-Stage Pharmaceutical Process Design

Problem Statement

hERG potassium channel blockade is the leading cause of drug withdrawal from the market and clinical trial termination due to cardiotoxicity. Current computational screening methods rely on 2D molecular descriptors and machine learning models that achieve AUROC values between 0.70 and 0.85 on benchmark datasets but fail to generalize across chemical scaffolds. The Chemical Process Systems programme at NSF funds research that develops new computational and experimental methods for designing safer chemical processes. I propose that topological data analysis, specifically persistent homology applied to drug-protein interaction graphs, captures higher-order structural features that 2D descriptors miss, enabling more accurate prediction of hERG blockade and earlier removal of toxic candidates from the drug development pipeline.

Proposed Approach

My platform TOPOLOGIX constructs bipartite simplicial complexes from drug-protein interaction data. Each complex encodes the binding geometry between a candidate molecule and the hERG channel pore domain. Persistent homology computes topological features across multiple scales: connected components (H0), cycles (H1), and voids (H2). These features form a topological fingerprint that I feed into a Bayesian logistic regression model. The preliminary MVP on 1,200 compounds from ChEMBL achieved an AUROC of 0.634. This is below the state of the art, but the model used only 50 topological features and a single protein conformation. The proposed work will expand the feature set to 200+ topological descriptors, incorporate ensemble docking across 10 hERG conformations from the Protein Data Bank, and train on 10,000+ compounds with known hERG IC50 values.

Validation Strategy

I will validate the model against three external test sets: the FDA-approved drug set (n=1,200), the Tox21 hERG dataset (n=800), and a set of 200 compounds withdrawn from the market for cardiotoxicity. The primary metric is AUROC, with a target of 0.85 or higher. Secondary metrics include sensitivity at 95% specificity and scaffold hold-out performance. I will compare TOPOLOGIX against four baselines: Morgan fingerprints with random forest, graph neural networks, 3D pharmacophore models, and the commercial tool ADMET Predictor.

Integration with Chemical Process Systems

The NSF Chemical Process Systems programme emphasizes the design of chemical processes that are safe, efficient, and sustainable. My framework integrates directly into pharmaceutical process design by providing a computational filter that removes cardiotoxic candidates before synthesis. This reduces material waste, animal testing, and clinical trial failures. The Bayesian framework quantifies prediction uncertainty, enabling risk-based decision-making in process development.

Timeline and Milestones

Months 1-3: Curate training dataset of 10,000 compounds from ChEMBL, PubChem, and Tox21. Compute 200 topological features per compound using Ripser and Gudhi. Months 4-6: Train Bayesian logistic regression model with PyMC. Validate on three external test sets. Months 7-9: Perform scaffold hold-out analysis and uncertainty quantification. Months 10-12: Write manuscript for submission to Journal of Chemical Information and Modeling or Chemical Engineering Science. Release TOPOLOGIX as open-source software under Apache 2.0 license.

Budget Justification

The requested funds support one year of computational infrastructure: cloud computing credits for high-throughput docking and persistent homology calculations (USD 15,000), software licenses for RDKit and Schrodinger (USD 5,000), publication fees for open-access journals (USD 3,000), and travel to one conference (USD 2,000). Total: USD 25,000. No salary support is requested. I hold a full-time position as National Product Manager at Synthcare in Lagos, Nigeria, which covers my living expenses.

CHECKLIST

- [ ] Confirm U.S. academic partner institution and obtain letter of support from principal investigator
- [ ] Register for NSF FastLane or Research.gov account
- [ ] Prepare project summary (one page, 4,600 characters max)
- [ ] Prepare project description (15 pages max, including references)
- [ ] Prepare biographical sketch for Eniola Ayodele Olutogun
- [ ] Prepare budget and budget justification (USD 25,000)
- [ ] Obtain current and pending support statement
- [ ] Submit through U.S. partner institution's sponsored research office
- [ ] Verify deadline on NSF PD 26-367Y programme website
- [ ] Confirm eligibility for non-U.S. citizen as subawardee or consultant

EDITOR NOTES

- Eligibility risk: The Chemical Process Systems programme requires the awardee to be a U.S. institution. Eniola is a Nigerian citizen with no current U.S. affiliation. The proposal must include a U.S. co-PI or host institution. Without this, the application will be returned without review. Identify potential collaborators at U.S. universities with computational chemistry or chemical engineering departments.
- Verification needed: The hERG cardiotoxicity MVP AUROC of 0.634 is stated in the profile but not published. Confirm the exact dataset size, compound source, and validation method before submitting. If the result is from an unpublished analysis, include a brief methods description in the project description.
- Gap: The profile does not specify which U.S. institution Eniola will partner with. The application materials assume a partner exists. Eniola must identify and contact at least one potential collaborator before submission. Suggested targets: University of Michigan (Berridge connection), Harvard (Gershman connection), or a chemical engineering department at MIT, Stanford, or University of Texas at Austin.
- Gap: No mention of prior NSF funding or familiarity with NSF proposal structure. Eniola should review the NSF Proposal and Award Policies and Procedures Guide (PAPPG) and consider using the NSF-funded Science and Technology Centers or Industry-University Cooperative Research Centers as potential partnership models.
- Tone adjustment: The motivation letter is direct and factual, but the NSF review panel may expect a clearer statement of broader impacts. Add one sentence on how safer pharmaceutical process design reduces healthcare costs and improves drug access in LMICs like Nigeria, without using the banned phrases.

AI Draft — Chemical Process Systems