Draft — Y Combinator · GrantFinder

MOTIVATION LETTER

Drug resistance kills drug development. A candidate antibiotic or anticancer agent that fails because of a single resistance mutation costs a company years and hundreds of millions of dollars. The tools to predict those mutations from structure cover only about 18 percent of known mutations, because they require a resolved protein structure. The other 82 percent are invisible to them.

TOPOLOGIX solves that problem. It predicts drug-resistance mutations from sequence alone, using ESM-2 protein language model delta-embeddings combined with Morgan drug fingerprints and a Random Forest classifier. On the Platinum benchmark of 553 mutations, it achieves an AUROC of 0.804 plus or minus 0.025. On SKEMPI 2.0, it reaches 0.634. It beats the structure-based baseline mCSM-lig, which scores around 0.70, while covering 100 percent of mutations instead of 18 percent. That is the proof of concept.

Y Combinator funds pre-seed startups that attack large, underserved markets with a defensible technical edge. The market here is concrete: every major pharmaceutical company runs resistance prediction as part of lead optimization, and every one of them is working from incomplete structural coverage. TOPOLOGIX is a software platform that plugs into their existing discovery pipelines as a SaaS product. The technical edge is the sequence-first approach, which removes the structure bottleneck entirely.

The founder background matches what YC looks for. I am a licensed pharmacist with a B.Pharm from the University of Ibadan, a computational modeler who has built and calibrated Bayesian ODE systems with PyMC, and a software engineer who has shipped production systems including Linux VPS operations, CI/CD pipelines, and automated backup infrastructure. I built TOPOLOGIX as an independent researcher, pre-registered the validation protocol, and reported the negative results from the earlier topology-based approach honestly before pivoting to sequence representations. That pivot is the product: the first approach failed its validation gate, and the second one beats the published baseline.

The path to a minimum viable product is short. The classifier already runs end to end. The next step is packaging it as an API with a web interface, then running a pilot with one or two pharma partners to generate the first revenue. Y Combinator's network and mentorship are the fastest route to those partnerships. The $500,000 investment funds the engineering time to productize the model, the compute for scaling to larger mutation databases, and the business development work to land the first customers.

TOPOLOGIX is a venture with a working model, a clear customer, and a founder who has already demonstrated the technical range to build it alone. YC's model of funding small teams with large technical use fits exactly what this company is.

RESEARCH STATEMENT

TOPOLOGIX predicts drug-resistance mutations from protein sequence data using a protein language model and drug fingerprint representation. The system takes a protein sequence and a drug molecule as input, and outputs a probability that a given mutation confers resistance to that drug. The current implementation uses ESM-2 delta-embeddings, which capture the change in the protein's learned representation when a mutation is introduced, concatenated with Morgan circular fingerprints of the drug. A Random Forest classifier maps that combined representation to a resistance label.

The validation results are the core of the technical claim. On the Platinum benchmark, which contains 553 mutations across multiple protein-drug systems, TOPOLOGIX achieves an AUROC of 0.804 with a standard deviation of 0.025. On SKEMPI 2.0, a binding affinity benchmark, it reaches 0.634. The structure-based tool mCSM-lig scores approximately 0.70 on comparable tasks, but it requires a resolved protein structure for every mutation it evaluates. That requirement excludes roughly 82 percent of known mutations. TOPOLOGIX covers all of them, because sequence data is available for every protein.

The development history is honest about what did not work. The first approach used bipartite persistent homology to characterize protein-ligand interface geometry, with an opposition-distance metric computed via Ripser and GUDHI. A pre-registered, powered replication on hERG cardiotoxicity showed that topological features did not beat a plain descriptor baseline: AUROC 0.8426 versus 0.8782. A second study applying the same topological constructs to drug-resistance prediction found almost no signal, with AUROC values of 0.425 and 0.485 on the Platinum benchmark. Those negative results ruled out interface geometry as the driver of resistance and motivated the sequence-representation approach that became TOPOLOGIX. The pivot is documented in pre-registered protocols, not retrofitted after the fact.

The technical roadmap has three stages. The first stage is productization: wrap the existing classifier in a REST API, build a simple web interface for submitting sequences and drugs, and add batch processing for whole-protein scanning. The second stage is data expansion: train on additional public resistance databases beyond Platinum and SKEMPI, including clinical resistance data where available, to improve generalization across protein families. The third stage is model improvement: replace the Random Forest with a gradient-boosted or deep learning model once the dataset is large enough, and add attention mechanisms to identify which residues drive the resistance prediction, giving medicinal chemists actionable information about which mutations to design around.

The scientific contribution is distinct from the commercial one. The finding that sequence-based representations outperform structure-based geometry for resistance prediction is a publishable result that challenges a common assumption in computational drug design. The finding that persistent homology features carry no signal for this task is also publishable, because it settles a comparison the literature had never actually run. Both results will be submitted to peer-reviewed venues as the product develops.

The commercial contribution is the product itself. Pharmaceutical companies run resistance screens during lead optimization, and they currently accept incomplete structural coverage as a limitation. TOPOLOGIX removes that limitation. The software is the deliverable, and the validation results are the evidence that it works.

ESSAY: WHY Y COMBINATOR

Y Combinator's model is built for a founder like me: a small team, a technical product, and a large market reached through software. TOPOLOGIX fits that model exactly. It is a single-founder company with a working classifier, a clear customer in pharmaceutical R&D, and a product that can be built and shipped by one engineer. YC's $500,000 investment and three-month program are the right size and shape for taking this from a validated model to a deployed product.

The specific value YC provides beyond capital is the network. TOPOLOGIX needs pilot customers in pharma, and YC's alumni network includes founders and operators who have sold software into pharmaceutical companies. The mentorship during the program, particularly around pricing and enterprise sales cycles, is directly relevant to a product that will be sold as a SaaS subscription to drug discovery teams. The demo day exposure to investors who understand biotech software is the natural fundraising path for the next round.

The timing is right. The model is validated. The market is underserved. The founder has the full technical stack in hand, from the ML pipeline to the production infrastructure. What is missing is the business development motion, and that is precisely what YC provides.

ESSAY: MARKET OPPORTUNITY

Drug resistance is a universal failure mode in drug development. Every antibiotic, every antifungal, every cancer therapy faces it. The companies that develop these drugs spend substantial resources trying to predict which mutations will emerge and design around them. The tools they use are structure-based, which means they fail for the majority of mutations that occur in proteins without resolved structures.

The market size is defined by the number of drug discovery programs that run resistance prediction. Each program spends on the order of hundreds of thousands of dollars per year on computational resistance analysis, whether through in-house teams or external software licenses. The global market for computational drug discovery tools is in the billions of dollars, and resistance prediction is a mandatory component of any serious lead optimization campaign.

TOPOLOGIX addresses this market with a clear differentiator: sequence-only input. That means it works for every protein, not just the 18 percent with structures. It also means it is faster, because sequence data is cheaper and easier to generate than structural data. The pricing model is a per-seat SaaS subscription with tiered access to batch processing and API integration. The first customers will be computational chemistry groups in mid-sized pharma and biotech companies, followed by large pharma once the tool is proven in production.

The competitive landscape is fragmented. Structure-based tools like mCSM-lig and its derivatives dominate, but they have the coverage problem. Machine learning approaches exist but are typically trained on narrow protein families and do not generalize. TOPOLOGIX is the only tool that combines protein language model representations with drug fingerprints in a generalizable, sequence-only framework. That is the defensible position.

ESSAY: TRACTION AND MILESTONES

The traction to date is the validation results. TOPOLOGIX achieves an AUROC of 0.804 on the Platinum benchmark, beating the structure-based baseline mCSM-lig at approximately 0.70 while covering 100 percent of mutations versus 18 percent. The model runs end to end on a standard laptop, and the code is version-controlled and reproducible.

The milestones for the next twelve months are specific. Month one: package the classifier as a REST API with a web interface. Month two: run a pilot with one external research group to test usability and gather feedback. Month three: incorporate feedback and add batch processing. Month four: approach two mid-sized pharma companies with a pilot proposal. Month six: sign the first paid pilot. Month nine: publish the sequence-versus-structure comparison in a peer-reviewed journal. Month twelve: convert the first pilot into a paid subscription and begin the next fundraising round.

The founder has the track record to execute these milestones. I built and calibrated a Bayesian ODE model of addiction neuropharmacology with 14 free parameters using PyMC, confirmed all five pre-registered hypotheses, and submitted three sole-authored preprints to peer-reviewed journals. I built a receptor-to-behavior brain-circuit simulation engine with 62 passing tests. I have shipped production systems including Linux VPS operations, CI/CD pipelines, and automated backup infrastructure. The gap is not technical execution; it is business development, and that is what YC provides.

CHECKLIST

- [ ] Complete Y Combinator online application at ycombinator.com/apply
- [ ] Submit company name: TOPOLOGIX
- [ ] Submit founder name: Eniola Ayodele Olutogun
- [ ] Submit founder background: B.Pharm, M.Sc. Digital Health (enrolled), software engineer
- [ ] Submit company description: sequence-based drug-resistance mutation prediction software
- [ ] Submit product URL or demo video link
- [ ] Submit GitHub repository link for TOPOLOGIX code
- [ ] Submit ORCID profile link
- [ ] Submit personal site link (zyco.org)
- [ ] Submit benchmark results summary (Platinum AUROC 0.804, SKEMPI 2.0 AUROC 0.634)
- [ ] Submit negative results documentation (hERG topology study, resistance topology study)
- [ ] Prepare 1-minute demo video showing model running on a sample mutation
- [ ] Prepare answers for YC interview questions on market size and pricing
- [ ] Verify eligibility for remote participation or plan for US relocation
- [ ] Confirm YC batch dates and application deadline for the target batch

EDITOR NOTES

- The chosen research line is TOPOLOGIX, consistent with the recommended framing angle. The negative topology results are presented as historical context for the pivot, not as current work, which matches the profile's honest reporting of falsified hypotheses.
- Eligibility risk: YC typically requires founders to relocate to the US for the program. Eniola is enrolled in an M.Sc. at HPI/Potsdam starting Winter 2026/27, which may conflict with a US-based accelerator. Verify YC's remote participation policy or plan for a leave of absence.
- The profile lists employment at Synthcare as National Product Manager from March 2026. Verify whether this is a full-time commitment that conflicts with YC's full-time founder requirement.
- The market size and pricing figures in the market opportunity essay are estimates. Eniola should replace them with specific numbers from market research reports or competitor pricing data before submission.
- The traction essay claims the model runs on a standard laptop. Verify this is true for the full Platinum benchmark, not just a subset, before stating it in the application.
- The YC application requires a demo video. This is not drafted here and must be produced by Eniola before submission.
- The profile does not specify whether TOPOLOGIX has any users or revenue. The essays correctly avoid claiming either. If any pilot conversations have occurred, add them to the traction essay.

MOTIVATION LETTER

Drug resistance kills drug development. A candidate antibiotic or anticancer agent that fails because of a single resistance mutation costs a company years and hundreds of millions of dollars. The tools to predict those mutations from structure cover only about 18 percent of known mutations, because they require a resolved protein structure. The other 82 percent are invisible to them.

TOPOLOGIX solves that problem. It predicts drug-resistance mutations from sequence alone, using ESM-2 protein language model delta-embeddings combined with Morgan drug fingerprints and a Random Forest classifier. On the Platinum benchmark of 553 mutations, it achieves an AUROC of 0.804 plus or minus 0.025. On SKEMPI 2.0, it reaches 0.634. It beats the structure-based baseline mCSM-lig, which scores around 0.70, while covering 100 percent of mutations instead of 18 percent. That is the proof of concept.

Y Combinator funds pre-seed startups that attack large, underserved markets with a defensible technical edge. The market here is concrete: every major pharmaceutical company runs resistance prediction as part of lead optimization, and every one of them is working from incomplete structural coverage. TOPOLOGIX is a software platform that plugs into their existing discovery pipelines as a SaaS product. The technical edge is the sequence-first approach, which removes the structure bottleneck entirely.

The founder background matches what YC looks for. I am a licensed pharmacist with a B.Pharm from the University of Ibadan, a computational modeler who has built and calibrated Bayesian ODE systems with PyMC, and a software engineer who has shipped production systems including Linux VPS operations, CI/CD pipelines, and automated backup infrastructure. I built TOPOLOGIX as an independent researcher, pre-registered the validation protocol, and reported the negative results from the earlier topology-based approach honestly before pivoting to sequence representations. That pivot is the product: the first approach failed its validation gate, and the second one beats the published baseline.

The path to a minimum viable product is short. The classifier already runs end to end. The next step is packaging it as an API with a web interface, then running a pilot with one or two pharma partners to generate the first revenue. Y Combinator's network and mentorship are the fastest route to those partnerships. The $500,000 investment funds the engineering time to productize the model, the compute for scaling to larger mutation databases, and the business development work to land the first customers.

TOPOLOGIX is a venture with a working model, a clear customer, and a founder who has already demonstrated the technical range to build it alone. YC's model of funding small teams with large technical use fits exactly what this company is.

RESEARCH STATEMENT

TOPOLOGIX predicts drug-resistance mutations from protein sequence data using a protein language model and drug fingerprint representation. The system takes a protein sequence and a drug molecule as input, and outputs a probability that a given mutation confers resistance to that drug. The current implementation uses ESM-2 delta-embeddings, which capture the change in the protein's learned representation when a mutation is introduced, concatenated with Morgan circular fingerprints of the drug. A Random Forest classifier maps that combined representation to a resistance label.

The validation results are the core of the technical claim. On the Platinum benchmark, which contains 553 mutations across multiple protein-drug systems, TOPOLOGIX achieves an AUROC of 0.804 with a standard deviation of 0.025. On SKEMPI 2.0, a binding affinity benchmark, it reaches 0.634. The structure-based tool mCSM-lig scores approximately 0.70 on comparable tasks, but it requires a resolved protein structure for every mutation it evaluates. That requirement excludes roughly 82 percent of known mutations. TOPOLOGIX covers all of them, because sequence data is available for every protein.

The development history is honest about what did not work. The first approach used bipartite persistent homology to characterize protein-ligand interface geometry, with an opposition-distance metric computed via Ripser and GUDHI. A pre-registered, powered replication on hERG cardiotoxicity showed that topological features did not beat a plain descriptor baseline: AUROC 0.8426 versus 0.8782. A second study applying the same topological constructs to drug-resistance prediction found almost no signal, with AUROC values of 0.425 and 0.485 on the Platinum benchmark. Those negative results ruled out interface geometry as the driver of resistance and motivated the sequence-representation approach that became TOPOLOGIX. The pivot is documented in pre-registered protocols, not retrofitted after the fact.

The technical roadmap has three stages. The first stage is productization: wrap the existing classifier in a REST API, build a simple web interface for submitting sequences and drugs, and add batch processing for whole-protein scanning. The second stage is data expansion: train on additional public resistance databases beyond Platinum and SKEMPI, including clinical resistance data where available, to improve generalization across protein families. The third stage is model improvement: replace the Random Forest with a gradient-boosted or deep learning model once the dataset is large enough, and add attention mechanisms to identify which residues drive the resistance prediction, giving medicinal chemists actionable information about which mutations to design around.

The scientific contribution is distinct from the commercial one. The finding that sequence-based representations outperform structure-based geometry for resistance prediction is a publishable result that challenges a common assumption in computational drug design. The finding that persistent homology features carry no signal for this task is also publishable, because it settles a comparison the literature had never actually run. Both results will be submitted to peer-reviewed venues as the product develops.

The commercial contribution is the product itself. Pharmaceutical companies run resistance screens during lead optimization, and they currently accept incomplete structural coverage as a limitation. TOPOLOGIX removes that limitation. The software is the deliverable, and the validation results are the evidence that it works.

ESSAY: WHY Y COMBINATOR

Y Combinator's model is built for a founder like me: a small team, a technical product, and a large market reached through software. TOPOLOGIX fits that model exactly. It is a single-founder company with a working classifier, a clear customer in pharmaceutical R&D, and a product that can be built and shipped by one engineer. YC's $500,000 investment and three-month program are the right size and shape for taking this from a validated model to a deployed product.

The specific value YC provides beyond capital is the network. TOPOLOGIX needs pilot customers in pharma, and YC's alumni network includes founders and operators who have sold software into pharmaceutical companies. The mentorship during the program, particularly around pricing and enterprise sales cycles, is directly relevant to a product that will be sold as a SaaS subscription to drug discovery teams. The demo day exposure to investors who understand biotech software is the natural fundraising path for the next round.

The timing is right. The model is validated. The market is underserved. The founder has the full technical stack in hand, from the ML pipeline to the production infrastructure. What is missing is the business development motion, and that is precisely what YC provides.

ESSAY: MARKET OPPORTUNITY

Drug resistance is a universal failure mode in drug development. Every antibiotic, every antifungal, every cancer therapy faces it. The companies that develop these drugs spend substantial resources trying to predict which mutations will emerge and design around them. The tools they use are structure-based, which means they fail for the majority of mutations that occur in proteins without resolved structures.

The market size is defined by the number of drug discovery programs that run resistance prediction. Each program spends on the order of hundreds of thousands of dollars per year on computational resistance analysis, whether through in-house teams or external software licenses. The global market for computational drug discovery tools is in the billions of dollars, and resistance prediction is a mandatory component of any serious lead optimization campaign.

TOPOLOGIX addresses this market with a clear differentiator: sequence-only input. That means it works for every protein, not just the 18 percent with structures. It also means it is faster, because sequence data is cheaper and easier to generate than structural data. The pricing model is a per-seat SaaS subscription with tiered access to batch processing and API integration. The first customers will be computational chemistry groups in mid-sized pharma and biotech companies, followed by large pharma once the tool is proven in production.

The competitive landscape is fragmented. Structure-based tools like mCSM-lig and its derivatives dominate, but they have the coverage problem. Machine learning approaches exist but are typically trained on narrow protein families and do not generalize. TOPOLOGIX is the only tool that combines protein language model representations with drug fingerprints in a generalizable, sequence-only framework. That is the defensible position.

ESSAY: TRACTION AND MILESTONES

The traction to date is the validation results. TOPOLOGIX achieves an AUROC of 0.804 on the Platinum benchmark, beating the structure-based baseline mCSM-lig at approximately 0.70 while covering 100 percent of mutations versus 18 percent. The model runs end to end on a standard laptop, and the code is version-controlled and reproducible.

The milestones for the next twelve months are specific. Month one: package the classifier as a REST API with a web interface. Month two: run a pilot with one external research group to test usability and gather feedback. Month three: incorporate feedback and add batch processing. Month four: approach two mid-sized pharma companies with a pilot proposal. Month six: sign the first paid pilot. Month nine: publish the sequence-versus-structure comparison in a peer-reviewed journal. Month twelve: convert the first pilot into a paid subscription and begin the next fundraising round.

The founder has the track record to execute these milestones. I built and calibrated a Bayesian ODE model of addiction neuropharmacology with 14 free parameters using PyMC, confirmed all five pre-registered hypotheses, and submitted three sole-authored preprints to peer-reviewed journals. I built a receptor-to-behavior brain-circuit simulation engine with 62 passing tests. I have shipped production systems including Linux VPS operations, CI/CD pipelines, and automated backup infrastructure. The gap is not technical execution; it is business development, and that is what YC provides.

CHECKLIST

- [ ] Complete Y Combinator online application at ycombinator.com/apply
- [ ] Submit company name: TOPOLOGIX
- [ ] Submit founder name: Eniola Ayodele Olutogun
- [ ] Submit founder background: B.Pharm, M.Sc. Digital Health (enrolled), software engineer
- [ ] Submit company description: sequence-based drug-resistance mutation prediction software
- [ ] Submit product URL or demo video link
- [ ] Submit GitHub repository link for TOPOLOGIX code
- [ ] Submit ORCID profile link
- [ ] Submit personal site link (zyco.org)
- [ ] Submit benchmark results summary (Platinum AUROC 0.804, SKEMPI 2.0 AUROC 0.634)
- [ ] Submit negative results documentation (hERG topology study, resistance topology study)
- [ ] Prepare 1-minute demo video showing model running on a sample mutation
- [ ] Prepare answers for YC interview questions on market size and pricing
- [ ] Verify eligibility for remote participation or plan for US relocation
- [ ] Confirm YC batch dates and application deadline for the target batch

EDITOR NOTES

- The chosen research line is TOPOLOGIX, consistent with the recommended framing angle. The negative topology results are presented as historical context for the pivot, not as current work, which matches the profile's honest reporting of falsified hypotheses.
- Eligibility risk: YC typically requires founders to relocate to the US for the program. Eniola is enrolled in an M.Sc. at HPI/Potsdam starting Winter 2026/27, which may conflict with a US-based accelerator. Verify YC's remote participation policy or plan for a leave of absence.
- The profile lists employment at Synthcare as National Product Manager from March 2026. Verify whether this is a full-time commitment that conflicts with YC's full-time founder requirement.
- The market size and pricing figures in the market opportunity essay are estimates. Eniola should replace them with specific numbers from market research reports or competitor pricing data before submission.
- The traction essay claims the model runs on a standard laptop. Verify this is true for the full Platinum benchmark, not just a subset, before stating it in the application.
- The YC application requires a demo video. This is not drafted here and must be produced by Eniola before submission.
- The profile does not specify whether TOPOLOGIX has any users or revenue. The essays correctly avoid claiming either. If any pilot conversations have occurred, add them to the traction essay.

AI Draft — Y Combinator