Draft — Futurize Incubator

MOTIVATION LETTER

The Platinum benchmark contains 553 mutations across 179 proteins, and structure-based tools like mCSM-lig can only score about 18 percent of them because they require a crystal structure. My platform, a sequence-only predictor of drug resistance mutations, covers 100 percent of those mutations and achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation. That is a measurable improvement over the published mCSM-lig result of roughly 0.70. The Futurize Incubator's healthtech mission, global reach, and pre-seed focus match exactly where this venture sits: a validated proof-of-concept that needs network access to biopharma partners and compute resources to fine-tune the underlying model on the SKEMPI 2.0 dataset of 3,000 mutation effects.

My background is pharmacy, not just machine learning. I hold a pharmacist degree and have worked as an ML engineer, which means I read a resistance mutation result with clinical context, not only statistical context. The venture predicts whether a given protein mutation will confer drug resistance using ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints, fed into a Random Forest classifier. No crystal structure is required. That design choice is deliberate: most clinically relevant mutations occur in proteins without solved structures, and the tools that depend on structures leave the majority of real-world cases unanswerable.

The Futurize Incubator's emphasis on digital health and AI-driven access aligns with the deployment model. A sequence-only predictor can run on modest infrastructure, which means it can serve laboratories and clinicians in regions where structural biology facilities do not exist. The same architecture that predicts resistance in oncology targets can be retrained for antiviral and antimicrobial resistance applications, which are urgent in African health systems. The proof-of-concept on Platinum and SKEMPI 2.0 is the technical foundation; the roadmap is fine-tuning ESM-2 on SKEMPI 3K mutations to reach an AUROC of at least 0.70, then running a pilot with Servier in Suresnes to convert the model into recurring revenue.

I am not incorporated yet, and I am a solo founder. Those are facts, not weaknesses. The venture is at pre-seed stage with a validated model, a clear partnership pipeline that includes Servier, Paris-Saclay I2BC, Institut Pasteur, and Sanofi in Gentilly, and a support pipeline that includes SEMIA, Quest for Health, WILCO One BioTech, IncubAlliance, and AI House. What I need from Futurize is the structured acceleration, the biopharma network, and the compute credits to move from benchmark validation to a commercial pilot. The model works. The next step is making it useful in a clinical workflow, and that is precisely the kind of translation Futurize exists to support.

RESEARCH STATEMENT

Drug resistance mutations are the primary reason targeted therapies fail, yet the standard computational tools for predicting them require a protein crystal structure. The Platinum benchmark, a curated set of 553 resistance-associated mutations across 179 proteins, shows the consequence: structure-dependent tools like mCSM-lig cover only about 18 percent of the dataset because the remaining 82 percent lack solved structures. My venture removes that dependency entirely. The method uses ESM-2 protein language model delta-embeddings, which capture the evolutionary and biophysical signal of a mutation from sequence alone, combined with ECFP4 drug fingerprints to represent the ligand. A Random Forest classifier then maps that joint representation to a resistance likelihood.

The validation results are specific. On Platinum, the model achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation, which prevents leakage between training and test mutations from the same protein. On SKEMPI 2.0, a broader mutation-effect benchmark, the AUROC is 0.634. The Platinum result beats the published state of the art for structure-based tools, mCSM-lig at roughly 0.70, while covering five times more mutations. The SKEMPI 2.0 result is lower and identifies the current bottleneck: the model needs fine-tuning on a larger, more diverse mutation-effect dataset to generalize beyond resistance-specific training data.

The roadmap is therefore concrete. First, fine-tune ESM-2 on the SKEMPI 3K mutation set, which contains thousands of experimentally measured binding affinity changes, to improve the model's general understanding of mutation effects on protein-ligand interactions. The target is an AUROC of at least 0.70 on SKEMPI 2.0 after fine-tuning. Second, run a pilot with Servier in Suresnes, where the model will be tested on internal oncology resistance cases to measure real-world clinical utility. Third, convert the pilot into an annual recurring revenue license for pharmaceutical companies and diagnostic labs.

The scientific rationale for sequence-only prediction is coverage, not convenience. Resistance mutations are frequently discovered in the clinic, in patients whose tumors or pathogens have evolved under drug pressure. Those mutations are often in proteins that have never been crystallized, or that crystallize poorly. A structure-based tool cannot answer the question at all. A sequence-based tool can, and it can be retrained for new protein families and new drug classes without waiting for structural biology. The same architecture applies to oncology resistance, antiviral resistance, and antimicrobial resistance, which is why the platform is positioned as a general drug resistance prediction engine rather than a single-disease tool.

The venture is at pre-seed stage. The proof-of-concept is validated on public benchmarks, but the company is not yet incorporated and has no revenue. The partnership pipeline is active: discussions with Servier, engagement with Paris-Saclay I2BC and Institut Pasteur for validation data, and a relationship with Sanofi in Gentilly. The immediate need is compute and mentorship to execute the SKEMPI fine-tuning and reach the pilot milestone. The model is the asset; the fine-tuning and pilot are the path to revenue.

ESSAY: FIT WITH FUTURIZE INCUBATOR

Futurize's stated focus is healthtech ventures that improve healthcare access and provider capacity, with a global geography that includes emerging markets. My venture fits that mission directly, not by analogy. A sequence-only resistance prediction platform removes the structural biology bottleneck that makes resistance testing inaccessible in most of the world. Crystal structure determination requires specialized facilities, expensive equipment, and trained personnel that simply do not exist in most African clinical and research settings. My model runs on a standard GPU and accepts a protein sequence, which any lab with sequencing capability already has. That is the same access logic that drives digital health platforms: extend capability beyond the clinic and the research center to wherever the patient and the pathogen are.

The incubator's pre-seed stage focus matches my current position. I am a solo founder with a validated proof-of-concept, not an incorporated company with revenue. I need exactly what an incubator provides at this stage: structured milestones, access to a biopharma network, and compute resources. The named partners in my pipeline, Servier, Paris-Saclay I2BC, Institut Pasteur, and Sanofi, are European, and Futurize's global network can help me convert those contacts into formal pilots. The compute credits are not a nice-to-have; fine-tuning ESM-2 on the SKEMPI 3K dataset is a real computational cost that I cannot currently absorb.

The founder fit is also direct. I am a pharmacist by training, which means I have clinical credibility when I talk to pharma partners about resistance. I am an ML engineer by practice, which means I built the model myself and can iterate on it without waiting for a technical cofounder. The incubator's typical criteria include founder commitment and coachability; I have committed to this venture as a sole founder and have already moved through SEMIA, Quest for Health, WILCO One BioTech, IncubAlliance, and AI House applications, which demonstrates a willingness to engage with structured support. The venture is early, the model is validated, and the mission is aligned with Futurize's global healthtech focus. That is the fit.

CHECKLIST

- [ ] Verify Futurize Incubator application deadline on programme website
- [ ] Confirm whether Futurize requires a pitch deck in addition to written materials
- [ ] Confirm whether Futurize requires a video pitch or founder introduction
- [ ] Confirm whether Futurize requires financial projections or a budget template
- [ ] Confirm whether Futurize requires a team section; if so, note solo founder status explicitly
- [ ] Verify current AUROC numbers against latest Platinum and SKEMPI 2.0 runs before submission
- [ ] Confirm Servier pilot discussion status and whether a letter of intent or support letter can be obtained
- [ ] Confirm whether incorporation is required before or during the incubator program
- [ ] Insert personal detail on why Futurize specifically, beyond generic healthtech alignment
- [ ] Insert personal detail on geographic focus, especially Nigeria or broader African deployment plans

EDITOR NOTES

- Eligibility risk: The strategy notes mention an e-pharmacy/telepharmacy platform for Nigeria, but the venture described in the applicant profile is a computational biology drug resistance prediction platform. These are different ventures. This draft assumes the computational biology venture is the correct one for Futurize based on the recommended framing angle, but the human reviewer must confirm which venture is actually being submitted before sending.
- Fact verification needed: The AUROC values (0.804 on Platinum, 0.634 on SKEMPI 2.0) and the mCSM-lig comparison (0.70) must be re-verified against the latest model runs before submission. The SKEMPI 3K fine-tuning target of AUROC 0.70 is a roadmap target, not a validated result, and must not be presented as achieved.
- Gap: The applicant profile does not specify a geographic focus for the computational biology venture. The motivation letter and essay reference African health systems and Nigeria implicitly, but the applicant must confirm whether that is the actual deployment target or whether the venture is global-first. If the venture is not Africa-focused, the Futurize fit argument weakens and must be revised.
- Gap: The applicant must insert a specific reason for choosing Futurize over the other incubators already in the pipeline (IncubAlliance, AI House, WILCO One BioTech). The current draft states alignment but does not distinguish Futurize from those alternatives.
- Risk: The venture is not incorporated and has no revenue. The application must not imply product-market fit or commercial traction. The current draft states pre-seed status honestly, but the applicant must ensure any verbal or pitch materials maintain the same discipline.

MOTIVATION LETTER

The Platinum benchmark contains 553 mutations across 179 proteins, and structure-based tools like mCSM-lig can only score about 18 percent of them because they require a crystal structure. My platform, a sequence-only predictor of drug resistance mutations, covers 100 percent of those mutations and achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation. That is a measurable improvement over the published mCSM-lig result of roughly 0.70. The Futurize Incubator's healthtech mission, global reach, and pre-seed focus match exactly where this venture sits: a validated proof-of-concept that needs network access to biopharma partners and compute resources to fine-tune the underlying model on the SKEMPI 2.0 dataset of 3,000 mutation effects.

My background is pharmacy, not just machine learning. I hold a pharmacist degree and have worked as an ML engineer, which means I read a resistance mutation result with clinical context, not only statistical context. The venture predicts whether a given protein mutation will confer drug resistance using ESM-2 protein language model delta-embeddings combined with ECFP4 drug fingerprints, fed into a Random Forest classifier. No crystal structure is required. That design choice is deliberate: most clinically relevant mutations occur in proteins without solved structures, and the tools that depend on structures leave the majority of real-world cases unanswerable.

The Futurize Incubator's emphasis on digital health and AI-driven access aligns with the deployment model. A sequence-only predictor can run on modest infrastructure, which means it can serve laboratories and clinicians in regions where structural biology facilities do not exist. The same architecture that predicts resistance in oncology targets can be retrained for antiviral and antimicrobial resistance applications, which are urgent in African health systems. The proof-of-concept on Platinum and SKEMPI 2.0 is the technical foundation; the roadmap is fine-tuning ESM-2 on SKEMPI 3K mutations to reach an AUROC of at least 0.70, then running a pilot with Servier in Suresnes to convert the model into recurring revenue.

I am not incorporated yet, and I am a solo founder. Those are facts, not weaknesses. The venture is at pre-seed stage with a validated model, a clear partnership pipeline that includes Servier, Paris-Saclay I2BC, Institut Pasteur, and Sanofi in Gentilly, and a support pipeline that includes SEMIA, Quest for Health, WILCO One BioTech, IncubAlliance, and AI House. What I need from Futurize is the structured acceleration, the biopharma network, and the compute credits to move from benchmark validation to a commercial pilot. The model works. The next step is making it useful in a clinical workflow, and that is precisely the kind of translation Futurize exists to support.

RESEARCH STATEMENT

Drug resistance mutations are the primary reason targeted therapies fail, yet the standard computational tools for predicting them require a protein crystal structure. The Platinum benchmark, a curated set of 553 resistance-associated mutations across 179 proteins, shows the consequence: structure-dependent tools like mCSM-lig cover only about 18 percent of the dataset because the remaining 82 percent lack solved structures. My venture removes that dependency entirely. The method uses ESM-2 protein language model delta-embeddings, which capture the evolutionary and biophysical signal of a mutation from sequence alone, combined with ECFP4 drug fingerprints to represent the ligand. A Random Forest classifier then maps that joint representation to a resistance likelihood.

The validation results are specific. On Platinum, the model achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation, which prevents leakage between training and test mutations from the same protein. On SKEMPI 2.0, a broader mutation-effect benchmark, the AUROC is 0.634. The Platinum result beats the published state of the art for structure-based tools, mCSM-lig at roughly 0.70, while covering five times more mutations. The SKEMPI 2.0 result is lower and identifies the current bottleneck: the model needs fine-tuning on a larger, more diverse mutation-effect dataset to generalize beyond resistance-specific training data.

The roadmap is therefore concrete. First, fine-tune ESM-2 on the SKEMPI 3K mutation set, which contains thousands of experimentally measured binding affinity changes, to improve the model's general understanding of mutation effects on protein-ligand interactions. The target is an AUROC of at least 0.70 on SKEMPI 2.0 after fine-tuning. Second, run a pilot with Servier in Suresnes, where the model will be tested on internal oncology resistance cases to measure real-world clinical utility. Third, convert the pilot into an annual recurring revenue license for pharmaceutical companies and diagnostic labs.

The scientific rationale for sequence-only prediction is coverage, not convenience. Resistance mutations are frequently discovered in the clinic, in patients whose tumors or pathogens have evolved under drug pressure. Those mutations are often in proteins that have never been crystallized, or that crystallize poorly. A structure-based tool cannot answer the question at all. A sequence-based tool can, and it can be retrained for new protein families and new drug classes without waiting for structural biology. The same architecture applies to oncology resistance, antiviral resistance, and antimicrobial resistance, which is why the platform is positioned as a general drug resistance prediction engine rather than a single-disease tool.

The venture is at pre-seed stage. The proof-of-concept is validated on public benchmarks, but the company is not yet incorporated and has no revenue. The partnership pipeline is active: discussions with Servier, engagement with Paris-Saclay I2BC and Institut Pasteur for validation data, and a relationship with Sanofi in Gentilly. The immediate need is compute and mentorship to execute the SKEMPI fine-tuning and reach the pilot milestone. The model is the asset; the fine-tuning and pilot are the path to revenue.

ESSAY: FIT WITH FUTURIZE INCUBATOR

Futurize's stated focus is healthtech ventures that improve healthcare access and provider capacity, with a global geography that includes emerging markets. My venture fits that mission directly, not by analogy. A sequence-only resistance prediction platform removes the structural biology bottleneck that makes resistance testing inaccessible in most of the world. Crystal structure determination requires specialized facilities, expensive equipment, and trained personnel that simply do not exist in most African clinical and research settings. My model runs on a standard GPU and accepts a protein sequence, which any lab with sequencing capability already has. That is the same access logic that drives digital health platforms: extend capability beyond the clinic and the research center to wherever the patient and the pathogen are.

The incubator's pre-seed stage focus matches my current position. I am a solo founder with a validated proof-of-concept, not an incorporated company with revenue. I need exactly what an incubator provides at this stage: structured milestones, access to a biopharma network, and compute resources. The named partners in my pipeline, Servier, Paris-Saclay I2BC, Institut Pasteur, and Sanofi, are European, and Futurize's global network can help me convert those contacts into formal pilots. The compute credits are not a nice-to-have; fine-tuning ESM-2 on the SKEMPI 3K dataset is a real computational cost that I cannot currently absorb.

The founder fit is also direct. I am a pharmacist by training, which means I have clinical credibility when I talk to pharma partners about resistance. I am an ML engineer by practice, which means I built the model myself and can iterate on it without waiting for a technical cofounder. The incubator's typical criteria include founder commitment and coachability; I have committed to this venture as a sole founder and have already moved through SEMIA, Quest for Health, WILCO One BioTech, IncubAlliance, and AI House applications, which demonstrates a willingness to engage with structured support. The venture is early, the model is validated, and the mission is aligned with Futurize's global healthtech focus. That is the fit.

CHECKLIST

- [ ] Verify Futurize Incubator application deadline on programme website
- [ ] Confirm whether Futurize requires a pitch deck in addition to written materials
- [ ] Confirm whether Futurize requires a video pitch or founder introduction
- [ ] Confirm whether Futurize requires financial projections or a budget template
- [ ] Confirm whether Futurize requires a team section; if so, note solo founder status explicitly
- [ ] Verify current AUROC numbers against latest Platinum and SKEMPI 2.0 runs before submission
- [ ] Confirm Servier pilot discussion status and whether a letter of intent or support letter can be obtained
- [ ] Confirm whether incorporation is required before or during the incubator program
- [ ] Insert personal detail on why Futurize specifically, beyond generic healthtech alignment
- [ ] Insert personal detail on geographic focus, especially Nigeria or broader African deployment plans

EDITOR NOTES

- Eligibility risk: The strategy notes mention an e-pharmacy/telepharmacy platform for Nigeria, but the venture described in the applicant profile is a computational biology drug resistance prediction platform. These are different ventures. This draft assumes the computational biology venture is the correct one for Futurize based on the recommended framing angle, but the human reviewer must confirm which venture is actually being submitted before sending.
- Fact verification needed: The AUROC values (0.804 on Platinum, 0.634 on SKEMPI 2.0) and the mCSM-lig comparison (0.70) must be re-verified against the latest model runs before submission. The SKEMPI 3K fine-tuning target of AUROC 0.70 is a roadmap target, not a validated result, and must not be presented as achieved.
- Gap: The applicant profile does not specify a geographic focus for the computational biology venture. The motivation letter and essay reference African health systems and Nigeria implicitly, but the applicant must confirm whether that is the actual deployment target or whether the venture is global-first. If the venture is not Africa-focused, the Futurize fit argument weakens and must be revised.
- Gap: The applicant must insert a specific reason for choosing Futurize over the other incubators already in the pipeline (IncubAlliance, AI House, WILCO One BioTech). The current draft states alignment but does not distinguish Futurize from those alternatives.
- Risk: The venture is not incorporated and has no revenue. The application must not imply product-market fit or commercial traction. The current draft states pre-seed status honestly, but the applicant must ensure any verbal or pitch materials maintain the same discipline.

AI Draft — Futurize Incubator