Draft — HealthX Incubation Program by Evon Labs

MOTIVATION LETTER

Antimicrobial resistance kills nearly five million people globally each year, and sub-Saharan Africa carries the highest burden per capita. The World Health Organization has declared AMR one of the top ten global public health threats. Yet the tools used to predict drug resistance mutations remain locked behind a structural biology bottleneck: they require crystal structures that exist for only about 18 percent of clinically relevant proteins. My venture removes that bottleneck entirely. The platform predicts drug resistance mutations from protein sequence alone, using ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints and a Random Forest classifier. On the Platinum benchmark, the model achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation, outperforming the published state of the art, mCSM-lig, which sits near 0.70. It also achieves 100 percent mutation coverage, compared to roughly 18 percent for structure-limited tools.

I am Eniola Olutogun, a pharmacist turned machine learning engineer, and I am applying to the HealthX Incubation Program by Evon Labs because this venture matches the program's focus on early-stage healthtech with scalable impact in African healthcare. The $30,000 grant and 12-week incubation would fund the next critical milestone: fine-tuning ESM-2 on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, which is the technical gate for a pilot with Servier in Suresnes. That pilot is the path to recurring revenue and the first commercial contract.

The Africa digital-health angle is central to my work. I am Nigerian, and my first market focus is Nigeria, where tuberculosis, HIV, and oncology programs struggle with rising drug resistance and where genomic surveillance infrastructure is thin. A tool that predicts resistance mutations from sequence data, without requiring crystal structures or high-end structural biology capacity, is a democratizing technology for low-resource laboratories. It turns a plain protein sequence into a resistance prediction in minutes.

I am committed to the full 12-week program. I have structured my roadmap so that the incubation period aligns with the model fine-tuning and validation work, not after it. I am open to mentorship and collaboration with other founders, particularly around regulatory pathways and clinical validation, which I have not yet navigated at this stage. The venture is pre-seed, proof-of-concept validated, and not yet incorporated, which means the program's support can shape the company's regulatory-aware infrastructure from day one.

HealthX selects founders who demonstrate commitment, openness, and a clear path to improving healthcare outcomes. I meet each criterion with evidence, not intention. The model works. The benchmark numbers are public and reproducible. The next step is a funded incubation to turn a validated proof of concept into a deployable product.

RESEARCH STATEMENT

The venture addresses a specific, measurable failure in computational drug discovery: predicting which mutations in a target protein confer resistance to a given drug, without requiring a crystal structure of that protein. Most existing tools, including mCSM-lig, depend on three-dimensional structural data. This dependency excludes the majority of clinically relevant proteins, for which no high-resolution structure exists. The result is that resistance prediction is unavailable precisely where it is needed most: emerging pathogens, orphan targets, and low-resource clinical settings.

My approach bypasses the structural bottleneck. The model takes a protein sequence and a drug fingerprint as inputs. The sequence is passed through ESM-2, a protein language model pre-trained on millions of natural sequences, and the delta-embedding between the wild-type and mutant sequence is computed. That delta-embedding is concatenated with the ECFP4 fingerprint of the drug. A Random Forest classifier then predicts whether the mutation confers resistance. No crystal structure is required at any point.

The validation results are concrete. On the Platinum benchmark, which contains 553 mutations and uses protein-grouped cross-validation to prevent data leakage, the model achieves an AUROC of 0.804 with a standard deviation of 0.025. This exceeds the published performance of mCSM-lig, which reports an AUROC around 0.70 on comparable tasks. On SKEMPI 2.0, a binding affinity benchmark, the model achieves an AUROC of 0.634, which is below the Platinum result but identifies a clear improvement target. The coverage advantage is stark: 100 percent of mutations in the benchmark can be scored, versus approximately 18 percent for structure-dependent tools.

The roadmap is specific. The next technical milestone is fine-tuning ESM-2 on the SKEMPI 3K mutation dataset, which contains roughly 3,000 experimentally measured mutation effects. The target is an AUROC of at least 0.70 on SKEMPI 2.0 after fine-tuning. This is the performance gate agreed with Servier in Suresnes for a pilot collaboration. The pilot would test the platform on a Servier-selected oncology target, with the goal of converting the pilot into a paid services contract and the first recurring revenue.

The commercial model is a software-as-a-service platform for biopharma resistance prediction, sold initially to mid-size European pharma companies, with a parallel path into African public health programs for AMR surveillance. The technical differentiator, sequence-only prediction, is also the market differentiator: it is faster, cheaper, and covers more targets than any structure-based competitor.

The scientific rationale is grounded in the biology of resistance. Resistance mutations cluster in functional regions of proteins, and their effects are encoded in the evolutionary and biophysical context of the sequence. A protein language model trained on millions of sequences captures evolutionary constraints that correlate with mutational tolerance. The delta-embedding isolates the change induced by a specific mutation. The drug fingerprint conditions the prediction on the specific molecule. This is a principled, mechanistically informed approach, not a black-box correlation.

The venture is at proof-of-concept stage. The model is validated on public benchmarks. It is not yet incorporated, has no revenue, and has not yet completed a commercial pilot. The HealthX Incubation Program is the right vehicle for the next phase because it provides non-dilutive capital, structured mentorship, and a network that includes biopharma partners. The $30,000 grant covers the compute and data engineering costs for the SKEMPI fine-tuning and the Servier pilot preparation. The 12-week structure matches the timeline from fine-tuning to pilot readiness.

ESSAY: SCALABILITY AND FEASIBILITY IN AFRICAN HEALTHCARE

The feasibility of this venture rests on a simple fact: the input is a protein sequence, and sequences are cheap and easy to obtain. Whole-genome and targeted sequencing capacity is expanding across Africa, including in Nigeria, where the National Agency for Food and Drug Administration and Control and the Nigeria Centre for Disease Control have both invested in genomic surveillance infrastructure. The bottleneck is not data generation; it is interpretation. My platform turns sequence data into resistance predictions without requiring structural biology expertise or expensive computational infrastructure. A laboratory with a standard server and a FASTA file can run the model.

Scalability in the African context follows from this. The platform is not tied to any single pathogen or drug class. The same architecture that predicts resistance in an oncology target for Servier can be applied to Mycobacterium tuberculosis, Plasmodium falciparum, or HIV protease. The model is retrained or fine-tuned per target class, but the core pipeline, sequence in, resistance prediction out, is unchanged. This means the platform can expand from oncology pilots into AMR surveillance programs across the continent without a fundamental redesign.

The HealthX program's emphasis on early-stage healthtech with African impact aligns with this venture's dual market strategy. The near-term revenue comes from European biopharma partnerships, starting with Servier. The long-term impact comes from deploying the platform in African public health programs, where drug resistance is a daily clinical reality. The $30,000 grant specifically funds the technical work that unlocks the Servier pilot, which is the first revenue milestone. The 12-week incubation provides the structured timeline and mentorship to convert a validated model into a company with a regulatory-aware infrastructure plan.

The commitment to the program is concrete. I have already initiated conversations with SEMIA and the Quest for Health program, and I have submitted applications to IncubAlliance and AI House in Paris. The HealthX program is not a speculative application; it is the next step in a planned sequence of support mechanisms, each tied to a specific milestone. The Servier pilot is the target, and the HealthX grant is the fuel.

CHECKLIST

- [ ] Confirm HealthX Incubation Program application deadline on the official programme website
- [ ] Verify that the $30,000 grant is non-dilutive and does not require equity
- [ ] Prepare a one-page executive summary of the venture for the application form
- [ ] Prepare a pitch deck (10-12 slides) covering problem, technology, validation, roadmap, and market
- [ ] Prepare a financial projection summary covering the 12-month runway from the grant
- [ ] Gather benchmark results documentation (Platinum AUROC 0.804, SKEMPI 2.0 AUROC 0.634)
- [ ] Prepare a one-page technical appendix describing ESM-2 delta-embeddings, ECFP4 fingerprints, and Random Forest architecture
- [ ] Draft a letter of intent or expression of interest from Servier for the pilot collaboration
- [ ] Confirm the status of the SEMIA / Quest for Health meeting and include as a reference if applicable
- [ ] Confirm the status of the IncubAlliance and AI House applications and include as context
- [ ] Prepare a short bio for Eniola Olutogun (pharmacist, ML engineer, sole founder)
- [ ] Confirm the legal status of the venture (pre-incorporation) and note any implications for the grant agreement
- [ ] Prepare a 12-week milestone plan aligned with the incubation program structure
- [ ] Identify two references who can speak to the technical validity of the model
- [ ] Review the application for any Nigeria-specific or Africa-specific requirements

EDITOR NOTES

- Eligibility risk: The programme targets early-stage healthtech founders; the venture is pre-incorporation, which may affect the grant agreement structure. Confirm whether the grant can be paid to an individual or requires an entity.
- Verification needed: The Servier pilot is described as a named partner, but the exact status of the collaboration (letter of intent, verbal agreement, or exploratory conversation) must be confirmed before submission. Do not overstate the commitment.
- Gap to fill: The application may ask for a specific founder bio or personal story. The profile does not include details on Eniola's pharmacy background, ML training path, or Nigeria-specific healthcare experience. Insert concrete personal details where the form requires them.
- The SKEMPI 2.0 AUROC of 0.634 is below the target of 0.70. This is honest but must be framed as a known improvement target with a specific plan (fine-tuning on SKEMPI 3K), not as a weakness.
- The programme's Africa digital-health angle is a strong fit, but the near-term revenue path is European pharma. Make sure the application does not over-index on Africa to the point of misrepresenting the commercial model.

MOTIVATION LETTER

Antimicrobial resistance kills nearly five million people globally each year, and sub-Saharan Africa carries the highest burden per capita. The World Health Organization has declared AMR one of the top ten global public health threats. Yet the tools used to predict drug resistance mutations remain locked behind a structural biology bottleneck: they require crystal structures that exist for only about 18 percent of clinically relevant proteins. My venture removes that bottleneck entirely. The platform predicts drug resistance mutations from protein sequence alone, using ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints and a Random Forest classifier. On the Platinum benchmark, the model achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation, outperforming the published state of the art, mCSM-lig, which sits near 0.70. It also achieves 100 percent mutation coverage, compared to roughly 18 percent for structure-limited tools.

I am Eniola Olutogun, a pharmacist turned machine learning engineer, and I am applying to the HealthX Incubation Program by Evon Labs because this venture matches the program's focus on early-stage healthtech with scalable impact in African healthcare. The $30,000 grant and 12-week incubation would fund the next critical milestone: fine-tuning ESM-2 on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, which is the technical gate for a pilot with Servier in Suresnes. That pilot is the path to recurring revenue and the first commercial contract.

The Africa digital-health angle is central to my work. I am Nigerian, and my first market focus is Nigeria, where tuberculosis, HIV, and oncology programs struggle with rising drug resistance and where genomic surveillance infrastructure is thin. A tool that predicts resistance mutations from sequence data, without requiring crystal structures or high-end structural biology capacity, is a democratizing technology for low-resource laboratories. It turns a plain protein sequence into a resistance prediction in minutes.

I am committed to the full 12-week program. I have structured my roadmap so that the incubation period aligns with the model fine-tuning and validation work, not after it. I am open to mentorship and collaboration with other founders, particularly around regulatory pathways and clinical validation, which I have not yet navigated at this stage. The venture is pre-seed, proof-of-concept validated, and not yet incorporated, which means the program's support can shape the company's regulatory-aware infrastructure from day one.

HealthX selects founders who demonstrate commitment, openness, and a clear path to improving healthcare outcomes. I meet each criterion with evidence, not intention. The model works. The benchmark numbers are public and reproducible. The next step is a funded incubation to turn a validated proof of concept into a deployable product.

RESEARCH STATEMENT

The venture addresses a specific, measurable failure in computational drug discovery: predicting which mutations in a target protein confer resistance to a given drug, without requiring a crystal structure of that protein. Most existing tools, including mCSM-lig, depend on three-dimensional structural data. This dependency excludes the majority of clinically relevant proteins, for which no high-resolution structure exists. The result is that resistance prediction is unavailable precisely where it is needed most: emerging pathogens, orphan targets, and low-resource clinical settings.

My approach bypasses the structural bottleneck. The model takes a protein sequence and a drug fingerprint as inputs. The sequence is passed through ESM-2, a protein language model pre-trained on millions of natural sequences, and the delta-embedding between the wild-type and mutant sequence is computed. That delta-embedding is concatenated with the ECFP4 fingerprint of the drug. A Random Forest classifier then predicts whether the mutation confers resistance. No crystal structure is required at any point.

The validation results are concrete. On the Platinum benchmark, which contains 553 mutations and uses protein-grouped cross-validation to prevent data leakage, the model achieves an AUROC of 0.804 with a standard deviation of 0.025. This exceeds the published performance of mCSM-lig, which reports an AUROC around 0.70 on comparable tasks. On SKEMPI 2.0, a binding affinity benchmark, the model achieves an AUROC of 0.634, which is below the Platinum result but identifies a clear improvement target. The coverage advantage is stark: 100 percent of mutations in the benchmark can be scored, versus approximately 18 percent for structure-dependent tools.

The roadmap is specific. The next technical milestone is fine-tuning ESM-2 on the SKEMPI 3K mutation dataset, which contains roughly 3,000 experimentally measured mutation effects. The target is an AUROC of at least 0.70 on SKEMPI 2.0 after fine-tuning. This is the performance gate agreed with Servier in Suresnes for a pilot collaboration. The pilot would test the platform on a Servier-selected oncology target, with the goal of converting the pilot into a paid services contract and the first recurring revenue.

The commercial model is a software-as-a-service platform for biopharma resistance prediction, sold initially to mid-size European pharma companies, with a parallel path into African public health programs for AMR surveillance. The technical differentiator, sequence-only prediction, is also the market differentiator: it is faster, cheaper, and covers more targets than any structure-based competitor.

The scientific rationale is grounded in the biology of resistance. Resistance mutations cluster in functional regions of proteins, and their effects are encoded in the evolutionary and biophysical context of the sequence. A protein language model trained on millions of sequences captures evolutionary constraints that correlate with mutational tolerance. The delta-embedding isolates the change induced by a specific mutation. The drug fingerprint conditions the prediction on the specific molecule. This is a principled, mechanistically informed approach, not a black-box correlation.

The venture is at proof-of-concept stage. The model is validated on public benchmarks. It is not yet incorporated, has no revenue, and has not yet completed a commercial pilot. The HealthX Incubation Program is the right vehicle for the next phase because it provides non-dilutive capital, structured mentorship, and a network that includes biopharma partners. The $30,000 grant covers the compute and data engineering costs for the SKEMPI fine-tuning and the Servier pilot preparation. The 12-week structure matches the timeline from fine-tuning to pilot readiness.

ESSAY: SCALABILITY AND FEASIBILITY IN AFRICAN HEALTHCARE

The feasibility of this venture rests on a simple fact: the input is a protein sequence, and sequences are cheap and easy to obtain. Whole-genome and targeted sequencing capacity is expanding across Africa, including in Nigeria, where the National Agency for Food and Drug Administration and Control and the Nigeria Centre for Disease Control have both invested in genomic surveillance infrastructure. The bottleneck is not data generation; it is interpretation. My platform turns sequence data into resistance predictions without requiring structural biology expertise or expensive computational infrastructure. A laboratory with a standard server and a FASTA file can run the model.

Scalability in the African context follows from this. The platform is not tied to any single pathogen or drug class. The same architecture that predicts resistance in an oncology target for Servier can be applied to Mycobacterium tuberculosis, Plasmodium falciparum, or HIV protease. The model is retrained or fine-tuned per target class, but the core pipeline, sequence in, resistance prediction out, is unchanged. This means the platform can expand from oncology pilots into AMR surveillance programs across the continent without a fundamental redesign.

The HealthX program's emphasis on early-stage healthtech with African impact aligns with this venture's dual market strategy. The near-term revenue comes from European biopharma partnerships, starting with Servier. The long-term impact comes from deploying the platform in African public health programs, where drug resistance is a daily clinical reality. The $30,000 grant specifically funds the technical work that unlocks the Servier pilot, which is the first revenue milestone. The 12-week incubation provides the structured timeline and mentorship to convert a validated model into a company with a regulatory-aware infrastructure plan.

The commitment to the program is concrete. I have already initiated conversations with SEMIA and the Quest for Health program, and I have submitted applications to IncubAlliance and AI House in Paris. The HealthX program is not a speculative application; it is the next step in a planned sequence of support mechanisms, each tied to a specific milestone. The Servier pilot is the target, and the HealthX grant is the fuel.

CHECKLIST

- [ ] Confirm HealthX Incubation Program application deadline on the official programme website
- [ ] Verify that the $30,000 grant is non-dilutive and does not require equity
- [ ] Prepare a one-page executive summary of the venture for the application form
- [ ] Prepare a pitch deck (10-12 slides) covering problem, technology, validation, roadmap, and market
- [ ] Prepare a financial projection summary covering the 12-month runway from the grant
- [ ] Gather benchmark results documentation (Platinum AUROC 0.804, SKEMPI 2.0 AUROC 0.634)
- [ ] Prepare a one-page technical appendix describing ESM-2 delta-embeddings, ECFP4 fingerprints, and Random Forest architecture
- [ ] Draft a letter of intent or expression of interest from Servier for the pilot collaboration
- [ ] Confirm the status of the SEMIA / Quest for Health meeting and include as a reference if applicable
- [ ] Confirm the status of the IncubAlliance and AI House applications and include as context
- [ ] Prepare a short bio for Eniola Olutogun (pharmacist, ML engineer, sole founder)
- [ ] Confirm the legal status of the venture (pre-incorporation) and note any implications for the grant agreement
- [ ] Prepare a 12-week milestone plan aligned with the incubation program structure
- [ ] Identify two references who can speak to the technical validity of the model
- [ ] Review the application for any Nigeria-specific or Africa-specific requirements

EDITOR NOTES

- Eligibility risk: The programme targets early-stage healthtech founders; the venture is pre-incorporation, which may affect the grant agreement structure. Confirm whether the grant can be paid to an individual or requires an entity.
- Verification needed: The Servier pilot is described as a named partner, but the exact status of the collaboration (letter of intent, verbal agreement, or exploratory conversation) must be confirmed before submission. Do not overstate the commitment.
- Gap to fill: The application may ask for a specific founder bio or personal story. The profile does not include details on Eniola's pharmacy background, ML training path, or Nigeria-specific healthcare experience. Insert concrete personal details where the form requires them.
- The SKEMPI 2.0 AUROC of 0.634 is below the target of 0.70. This is honest but must be framed as a known improvement target with a specific plan (fine-tuning on SKEMPI 3K), not as a weakness.
- The programme's Africa digital-health angle is a strong fit, but the near-term revenue path is European pharma. Make sure the application does not over-index on Africa to the point of misrepresenting the commercial model.

AI Draft — HealthX Incubation Program by Evon Labs