Draft — Africa Health-Tech Accelerator 2026

MOTIVATION LETTER

The Africa Health-Tech Accelerator 2026 selects ventures where data and intelligent technologies meet measurable clinical need. My venture predicts drug resistance mutations from protein sequence alone, using a protein language model and drug fingerprint ensemble that achieves an AUROC of 0.804 on the Platinum benchmark. That benchmark covers 553 mutations with protein-grouped cross-validation, and my model reaches 100 percent mutation coverage, where structure-limited tools cover roughly 18 percent. The published state of the art, mCSM-lig, sits near 0.70 AUROC on comparable tasks. My approach beats it without requiring a crystal structure, which means it works on proteins that have never been crystallized, including many resistance targets relevant to African pathogens.

The venture is pre-seed, with a validated proof of concept and a clear technical roadmap. I am a pharmacist turned machine learning engineer, and I built the system as a sole author. The next step is fine-tuning the ESM-2 backbone on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, then piloting with Servier in Suresnes. I have named research partners at Paris-Saclay, I2BC, and Institut Pasteur, and I am in active discussions with SEMIA and Quest for Health. I have submitted applications to IncubAlliance and AI House, with EIC Accelerator and BPI i-Lab as near-term targets.

The Africa Health-Tech Accelerator fits this venture for three concrete reasons. First, the programme prioritizes data and intelligent technologies in health innovation, and my model is exactly that: a sequence-only predictor that removes the structural biology bottleneck from resistance forecasting. Second, the programme requires a minimum viable product, and I have one, validated on public benchmarks with reproducible cross-validation. Third, the programme emphasizes scaling across Africa, and antimicrobial resistance is a documented priority for the continent. My Nigerian background and my regulatory-aware approach, built from pharmacy training in a Nigeria-first e-pharmacy context, position me to understand deployment constraints that a purely European team would miss.

I am applying as a solo founder, which I know falls short of the programme's two-founder minimum. My plan is to bring on a co-founder with complementary expertise in clinical microbiology or infectious disease before the programme start date. I have already identified two candidates through my Institut Pasteur network and am in active conversations. I am not asking for an exception; I am stating the timeline and the commitment.

The venture is not incorporated yet, but the proof of concept is complete, the benchmark numbers are public, and the compute path is clear. The Africa Health-Tech Accelerator's ecosystem access, investor network, and focus on investment-ready ventures match the stage I am at and the milestones I need to hit. I am ready to present the technical work, the validation data, and the Africa-specific use cases in detail.

RESEARCH STATEMENT

Drug resistance is a prediction problem before it is a treatment problem. When a pathogen acquires a mutation that evades a drug, clinicians lose time and patients lose options. My venture addresses the earliest point in that chain: predicting which mutations will confer resistance, from the protein sequence alone, before the mutation appears in a clinical isolate.

The technical core is a Random Forest classifier trained on delta embeddings from the ESM-2 protein language model, concatenated with ECFP4 drug fingerprints. This design captures two complementary signals. The ESM-2 delta embeddings represent the evolutionary and structural context of a mutation as encoded in language model space, without requiring a crystal structure. The ECFP4 fingerprints represent the chemical identity of the drug being tested. The classifier learns the interaction between those two representations.

Validation results are specific and reproducible. On the Platinum benchmark, which contains 553 mutations across diverse protein targets, the model achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation. This grouping is critical: it ensures the model is tested on proteins it has never seen during training, which is the realistic deployment scenario. On SKEMPI 2.0, a binding affinity benchmark, the model achieves 0.634 AUROC. That number is lower, but it reflects a harder transfer task, and it is the target for the next fine-tuning round.

The comparison to existing tools is stark. Structure-limited methods like mCSM-lig require a resolved crystal structure for the target protein. For many resistance-relevant proteins, especially those from neglected pathogens, no such structure exists. My model covers 100 percent of mutations in the Platinum benchmark because it only needs sequence. Structure-limited tools cover approximately 18 percent. That is a change in what problems can be addressed at all, not an incremental improvement.

The roadmap has three phases. Phase one, currently underway, is fine-tuning the ESM-2 backbone on the SKEMPI 3K mutation dataset, with a target AUROC of at least 0.70 on that benchmark. Phase two is a pilot with Servier in Suresnes, applying the model to their internal resistance questions and converting the benchmark performance into a commercial reference. Phase three is building annual recurring revenue around a software-as-a-service model for biopharma discovery teams.

The Africa angle is not an afterthought. Antimicrobial resistance rates in African health systems are among the highest globally, and the pathogen genomes circulating there are under-sequenced relative to their burden. A sequence-only predictor is the only kind that can scale to that context, because it does not depend on structural biology infrastructure that does not exist in most African research settings. The model can be retrained on African genomic surveillance data as it becomes available, and the same architecture that predicts drug resistance mutations can be adapted to predict vaccine escape mutations.

The venture is pre-seed, sole-founder, and not yet incorporated. The science is validated. The next twelve months are about turning that validation into a pilot, a team, and a revenue line.

SHORT ESSAY: SCALING ACROSS AFRICA

Antimicrobial resistance is a current clinical reality in Africa, not a future threat. My model predicts resistance mutations from protein sequence alone, which matters specifically because African pathogen genomes are under-sequenced and under-annotated. Structure-based tools fail on these targets because the crystal structures do not exist. My approach does not need them.

The venture's Africa strategy starts with data. As African genomic surveillance programs publish more sequences, the model can be fine-tuned on those sequences to reflect regional resistance patterns. The architecture is already validated on the Platinum benchmark at 0.804 AUROC, and the same pipeline can ingest new sequence data without re-engineering.

The second layer is regulatory awareness. My pharmacy training and my work on a Nigeria-first e-pharmacy gave me direct experience with Nigerian health regulatory pathways. That is a deployment requirement, not a side skill. Any tool that informs drug choice or resistance monitoring in Nigeria will need NAFDAC alignment, and I understand that process from the inside.

The third layer is partnership. The Africa Health-Tech Accelerator's network across the continent is the entry point for hospital system pilots and public health collaborations. I am not looking for a distribution channel; I am looking for clinical validation sites where the model can be tested against real resistance data from African isolates.

The venture is early, but the problem is not. AMR is projected to cause millions of deaths annually in Africa within the next two decades. A sequence-only predictor is the most scalable tool for staying ahead of that curve.

SHORT ESSAY: INVESTMENT READINESS AND MVP

The minimum viable product is complete and validated. The model is a Random Forest classifier trained on ESM-2 delta embeddings and ECFP4 drug fingerprints. It achieves 0.804 AUROC on the Platinum benchmark with protein-grouped cross-validation, and 0.634 on SKEMPI 2.0. It covers 100 percent of mutations in the benchmark, where structure-limited tools cover about 18 percent. These are public, reproducible numbers.

The venture is pre-seed and not yet incorporated, which I address head-on. The technical risk is retired; the commercial risk is what remains. The next milestone is a pilot with Servier in Suresnes, which requires incorporation and a formal data-sharing agreement. I have named research partners at Paris-Saclay, I2BC, and Institut Pasteur, and I am in active discussions with SEMIA and Quest for Health.

The two-founder requirement is a known gap. I am a sole founder and I am actively recruiting a co-founder with clinical microbiology or infectious disease expertise. I have two candidates identified through my Institut Pasteur network. My plan is to have a co-founder on board before the programme start date, not as a formality but as a working partner who owns the clinical validation workstream.

The business model is software-as-a-service for biopharma discovery teams, with the Servier pilot as the first reference customer. The Africa Health-Tech Accelerator's focus on investment-ready ventures matches this stage: the product works, the market is clear, and the next capital should go toward pilot execution and team building.

CHECKLIST

- [ ] Confirm the Africa Health-Tech Accelerator 2026 application deadline on the programme website
- [ ] Verify whether the two-founder minimum is a hard eligibility gate or a scoring criterion
- [ ] Secure a co-founder commitment letter or a signed co-founder agreement before submission
- [ ] Prepare a pitch deck with benchmark results, model architecture diagram, and roadmap slides
- [ ] Prepare a one-page business plan summary with revenue model and pilot timeline
- [ ] Include the Servier pilot LOI or a letter of intent from a named research partner
- [ ] Include the Platinum benchmark AUROC 0.804 and SKEMPI 2.0 AUROC 0.634 results as an appendix
- [ ] Confirm incorporation timeline for France (SAS or SASU) and include in the application
- [ ] Draft a data-sharing and IP ownership statement for the Servier pilot
- [ ] Prepare a one-page Africa market analysis focused on AMR burden and genomic surveillance gaps
- [ ] Submit the application through the startupmapafrica.com portal
- [ ] Follow up with the programme team by email within five business days of submission

EDITOR NOTES

- Eligibility risk: the two-founder minimum is stated in the selection criteria. The application must include a concrete co-founder plan, not just an intention. If a commitment letter cannot be secured before submission, flag this in the application narrative and provide a dated recruitment timeline.
- Fact verification: confirm the Platinum benchmark AUROC of 0.804 and SKEMPI 2.0 AUROC of 0.634 are from the latest model version and that the protein-grouped cross-validation description is accurate. Confirm the mCSM-lig AUROC of 0.70 is the correct published comparison point.
- Gap: the profile does not specify the exact deadline for the Africa Health-Tech Accelerator 2026. The applicant must check the programme website and work backward from that date.
- Gap: the profile does not name the two co-founder candidates. The applicant must insert names, affiliations, and the specific expertise each would bring.
- Gap: the profile does not specify the e-pharmacy venture name or the regulatory body (NAFDAC) experience in detail. The applicant should add one or two concrete examples of regulatory work to strengthen the Africa credibility claim.

MOTIVATION LETTER

The Africa Health-Tech Accelerator 2026 selects ventures where data and intelligent technologies meet measurable clinical need. My venture predicts drug resistance mutations from protein sequence alone, using a protein language model and drug fingerprint ensemble that achieves an AUROC of 0.804 on the Platinum benchmark. That benchmark covers 553 mutations with protein-grouped cross-validation, and my model reaches 100 percent mutation coverage, where structure-limited tools cover roughly 18 percent. The published state of the art, mCSM-lig, sits near 0.70 AUROC on comparable tasks. My approach beats it without requiring a crystal structure, which means it works on proteins that have never been crystallized, including many resistance targets relevant to African pathogens.

The venture is pre-seed, with a validated proof of concept and a clear technical roadmap. I am a pharmacist turned machine learning engineer, and I built the system as a sole author. The next step is fine-tuning the ESM-2 backbone on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, then piloting with Servier in Suresnes. I have named research partners at Paris-Saclay, I2BC, and Institut Pasteur, and I am in active discussions with SEMIA and Quest for Health. I have submitted applications to IncubAlliance and AI House, with EIC Accelerator and BPI i-Lab as near-term targets.

The Africa Health-Tech Accelerator fits this venture for three concrete reasons. First, the programme prioritizes data and intelligent technologies in health innovation, and my model is exactly that: a sequence-only predictor that removes the structural biology bottleneck from resistance forecasting. Second, the programme requires a minimum viable product, and I have one, validated on public benchmarks with reproducible cross-validation. Third, the programme emphasizes scaling across Africa, and antimicrobial resistance is a documented priority for the continent. My Nigerian background and my regulatory-aware approach, built from pharmacy training in a Nigeria-first e-pharmacy context, position me to understand deployment constraints that a purely European team would miss.

I am applying as a solo founder, which I know falls short of the programme's two-founder minimum. My plan is to bring on a co-founder with complementary expertise in clinical microbiology or infectious disease before the programme start date. I have already identified two candidates through my Institut Pasteur network and am in active conversations. I am not asking for an exception; I am stating the timeline and the commitment.

The venture is not incorporated yet, but the proof of concept is complete, the benchmark numbers are public, and the compute path is clear. The Africa Health-Tech Accelerator's ecosystem access, investor network, and focus on investment-ready ventures match the stage I am at and the milestones I need to hit. I am ready to present the technical work, the validation data, and the Africa-specific use cases in detail.

RESEARCH STATEMENT

Drug resistance is a prediction problem before it is a treatment problem. When a pathogen acquires a mutation that evades a drug, clinicians lose time and patients lose options. My venture addresses the earliest point in that chain: predicting which mutations will confer resistance, from the protein sequence alone, before the mutation appears in a clinical isolate.

The technical core is a Random Forest classifier trained on delta embeddings from the ESM-2 protein language model, concatenated with ECFP4 drug fingerprints. This design captures two complementary signals. The ESM-2 delta embeddings represent the evolutionary and structural context of a mutation as encoded in language model space, without requiring a crystal structure. The ECFP4 fingerprints represent the chemical identity of the drug being tested. The classifier learns the interaction between those two representations.

Validation results are specific and reproducible. On the Platinum benchmark, which contains 553 mutations across diverse protein targets, the model achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation. This grouping is critical: it ensures the model is tested on proteins it has never seen during training, which is the realistic deployment scenario. On SKEMPI 2.0, a binding affinity benchmark, the model achieves 0.634 AUROC. That number is lower, but it reflects a harder transfer task, and it is the target for the next fine-tuning round.

The comparison to existing tools is stark. Structure-limited methods like mCSM-lig require a resolved crystal structure for the target protein. For many resistance-relevant proteins, especially those from neglected pathogens, no such structure exists. My model covers 100 percent of mutations in the Platinum benchmark because it only needs sequence. Structure-limited tools cover approximately 18 percent. That is a change in what problems can be addressed at all, not an incremental improvement.

The roadmap has three phases. Phase one, currently underway, is fine-tuning the ESM-2 backbone on the SKEMPI 3K mutation dataset, with a target AUROC of at least 0.70 on that benchmark. Phase two is a pilot with Servier in Suresnes, applying the model to their internal resistance questions and converting the benchmark performance into a commercial reference. Phase three is building annual recurring revenue around a software-as-a-service model for biopharma discovery teams.

The Africa angle is not an afterthought. Antimicrobial resistance rates in African health systems are among the highest globally, and the pathogen genomes circulating there are under-sequenced relative to their burden. A sequence-only predictor is the only kind that can scale to that context, because it does not depend on structural biology infrastructure that does not exist in most African research settings. The model can be retrained on African genomic surveillance data as it becomes available, and the same architecture that predicts drug resistance mutations can be adapted to predict vaccine escape mutations.

The venture is pre-seed, sole-founder, and not yet incorporated. The science is validated. The next twelve months are about turning that validation into a pilot, a team, and a revenue line.

SHORT ESSAY: SCALING ACROSS AFRICA

Antimicrobial resistance is a current clinical reality in Africa, not a future threat. My model predicts resistance mutations from protein sequence alone, which matters specifically because African pathogen genomes are under-sequenced and under-annotated. Structure-based tools fail on these targets because the crystal structures do not exist. My approach does not need them.

The venture's Africa strategy starts with data. As African genomic surveillance programs publish more sequences, the model can be fine-tuned on those sequences to reflect regional resistance patterns. The architecture is already validated on the Platinum benchmark at 0.804 AUROC, and the same pipeline can ingest new sequence data without re-engineering.

The second layer is regulatory awareness. My pharmacy training and my work on a Nigeria-first e-pharmacy gave me direct experience with Nigerian health regulatory pathways. That is a deployment requirement, not a side skill. Any tool that informs drug choice or resistance monitoring in Nigeria will need NAFDAC alignment, and I understand that process from the inside.

The third layer is partnership. The Africa Health-Tech Accelerator's network across the continent is the entry point for hospital system pilots and public health collaborations. I am not looking for a distribution channel; I am looking for clinical validation sites where the model can be tested against real resistance data from African isolates.

The venture is early, but the problem is not. AMR is projected to cause millions of deaths annually in Africa within the next two decades. A sequence-only predictor is the most scalable tool for staying ahead of that curve.

SHORT ESSAY: INVESTMENT READINESS AND MVP

The minimum viable product is complete and validated. The model is a Random Forest classifier trained on ESM-2 delta embeddings and ECFP4 drug fingerprints. It achieves 0.804 AUROC on the Platinum benchmark with protein-grouped cross-validation, and 0.634 on SKEMPI 2.0. It covers 100 percent of mutations in the benchmark, where structure-limited tools cover about 18 percent. These are public, reproducible numbers.

The venture is pre-seed and not yet incorporated, which I address head-on. The technical risk is retired; the commercial risk is what remains. The next milestone is a pilot with Servier in Suresnes, which requires incorporation and a formal data-sharing agreement. I have named research partners at Paris-Saclay, I2BC, and Institut Pasteur, and I am in active discussions with SEMIA and Quest for Health.

The two-founder requirement is a known gap. I am a sole founder and I am actively recruiting a co-founder with clinical microbiology or infectious disease expertise. I have two candidates identified through my Institut Pasteur network. My plan is to have a co-founder on board before the programme start date, not as a formality but as a working partner who owns the clinical validation workstream.

The business model is software-as-a-service for biopharma discovery teams, with the Servier pilot as the first reference customer. The Africa Health-Tech Accelerator's focus on investment-ready ventures matches this stage: the product works, the market is clear, and the next capital should go toward pilot execution and team building.

CHECKLIST

- [ ] Confirm the Africa Health-Tech Accelerator 2026 application deadline on the programme website
- [ ] Verify whether the two-founder minimum is a hard eligibility gate or a scoring criterion
- [ ] Secure a co-founder commitment letter or a signed co-founder agreement before submission
- [ ] Prepare a pitch deck with benchmark results, model architecture diagram, and roadmap slides
- [ ] Prepare a one-page business plan summary with revenue model and pilot timeline
- [ ] Include the Servier pilot LOI or a letter of intent from a named research partner
- [ ] Include the Platinum benchmark AUROC 0.804 and SKEMPI 2.0 AUROC 0.634 results as an appendix
- [ ] Confirm incorporation timeline for France (SAS or SASU) and include in the application
- [ ] Draft a data-sharing and IP ownership statement for the Servier pilot
- [ ] Prepare a one-page Africa market analysis focused on AMR burden and genomic surveillance gaps
- [ ] Submit the application through the startupmapafrica.com portal
- [ ] Follow up with the programme team by email within five business days of submission

EDITOR NOTES

- Eligibility risk: the two-founder minimum is stated in the selection criteria. The application must include a concrete co-founder plan, not just an intention. If a commitment letter cannot be secured before submission, flag this in the application narrative and provide a dated recruitment timeline.
- Fact verification: confirm the Platinum benchmark AUROC of 0.804 and SKEMPI 2.0 AUROC of 0.634 are from the latest model version and that the protein-grouped cross-validation description is accurate. Confirm the mCSM-lig AUROC of 0.70 is the correct published comparison point.
- Gap: the profile does not specify the exact deadline for the Africa Health-Tech Accelerator 2026. The applicant must check the programme website and work backward from that date.
- Gap: the profile does not name the two co-founder candidates. The applicant must insert names, affiliations, and the specific expertise each would bring.
- Gap: the profile does not specify the e-pharmacy venture name or the regulatory body (NAFDAC) experience in detail. The applicant should add one or two concrete examples of regulatory work to strengthen the Africa credibility claim.

AI Draft — Africa Health-Tech Accelerator 2026