Draft — L’Oréal Green Sciences Incubator Startup Call 2026

MOTIVATION LETTER

The L’Oréal Green Sciences Incubator Startup Call 2026 targets ventures that convert renewable carbon into valuable molecules with measurable environmental benefit. My venture does exactly that, but with a computational twist that eliminates the most resource-intensive step in molecular design: experimental resistance screening. I am Eniola Olutogun, a pharmacist and machine learning engineer, and I have built a platform that predicts drug resistance mutations from protein sequence alone, without requiring crystal structures. The model achieves an AUROC of 0.804 on the Platinum benchmark across 553 mutations, with protein-grouped cross-validation, and covers 100 percent of mutations tested versus roughly 18 percent for structure-dependent tools. The proof-of-concept is validated and reproducible.

The green sciences connection is direct. Drug discovery today burns enormous resources on high-throughput experimental screening to identify which candidate molecules will fail due to resistance. Each failed candidate represents embodied carbon, chemical waste, and energy spent on synthesis and assay. My platform predicts resistance in silico, so researchers can discard weak candidates before they reach the lab. For antimicrobial and antiviral programs, this means fewer synthesis cycles, less solvent waste, and a shorter path to molecules that persist in efficacy. That is renewable carbon stewardship applied to pharmaceutical R&D.

I am applying to this specific incubator because Genopole’s facilities and L’Oréal’s innovation network offer what a pre-seed computational venture needs most: wet-lab partners to validate predictions and an industrial lens on bio-based molecule performance. The four domains of this call, particularly renewable carbon into valuable molecules, match my roadmap. I plan to fine-tune the ESM-2 protein language model on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, then run a pilot with Servier in Suresnes to demonstrate utility in an industrial resistance workflow. The incubator’s one-year on-site term at Genopole aligns with that timeline.

I also see a concrete bridge to cosmetics. Preservative efficacy and microbial resistance to bio-based actives are real problems in formulation science. My model can predict whether microbes will evolve resistance to antimicrobial peptides or preservative systems, which is directly relevant to L’Oréal’s green sciences agenda. The same architecture that predicts drug resistance mutations can be retrained on preservative and active ingredient datasets. This is the same underlying biology of microbial adaptation applied to a different molecular class.

The venture is pre-seed, not yet incorporated, and I am the sole founder. The model is built and benchmarked. What I need from this incubator is the industrial network, the wet-lab validation capacity, and the structured support to move from proof-of-concept to pilot revenue. I am asking for a seat in the 2026 cohort to do that work at Genopole.

RESEARCH STATEMENT

The scientific problem is precise: drug resistance emerges through specific mutations in protein targets, and current prediction tools require crystal structures that exist for only a fraction of clinically relevant proteins. My venture removes that requirement. The method uses ESM-2 protein language model delta-embeddings, which capture the evolutionary and structural context of a mutation from sequence alone, combined with ECFP4 drug fingerprints to represent the ligand. A Random Forest classifier then predicts whether a given mutation confers resistance to a given drug. This architecture is deliberately simple, interpretable, and fast enough to screen entire proteomes in hours.

Validation status is honest and specific. On the Platinum benchmark, which contains 553 mutations across diverse protein-drug pairs, 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 reflects real-world deployment. On SKEMPI 2.0, a harder benchmark with more sparse mutation data, the AUROC is 0.634. That number is the current ceiling and the target for improvement. Published state-of-the-art tools such as mCSM-lig achieve roughly 0.70 AUROC on similar tasks, but they require crystal structures. My model matches or exceeds that performance while covering 100 percent of mutations in the benchmark, versus approximately 18 percent for structure-limited tools. Coverage is the difference between a tool that works in research and a tool that works in production.

The roadmap has three stages. First, fine-tune ESM-2 on the SKEMPI 3K mutation dataset, which contains roughly 3,000 mutation-effect measurements. The goal is to raise SKEMPI 2.0 AUROC from 0.634 to at least 0.70. This is an engineering task with a clear success metric. Second, run a pilot with Servier in Suresnes to apply the model to an active drug discovery program, specifically to pre-screen candidate molecules for resistance liability before experimental validation. The pilot has a defined deliverable: a ranked list of resistance-prone mutations for a Servier target, with experimental confirmation of at least the top five predictions. Third, convert the pilot into a paid subscription or per-project license, establishing the venture’s first recurring revenue.

The green sciences relevance is embedded in the method itself. Every mutation prediction that is correct in silico is an experiment not run in the lab. Each avoided experiment saves reagents, solvents, energy, and labor. For a typical antimicrobial discovery program running 10,000 compound screens per year, even a 20 percent reduction in failed candidates translates to measurable reductions in chemical waste and energy consumption. The platform also enables the design of bio-based therapeutics, which are synthesized from renewable feedstocks, by ensuring those molecules are not abandoned early due to uncharacterized resistance risk. This is renewable carbon into valuable molecules, with the computational layer as the efficiency engine.

The fit with L’Oréal’s Green Sciences Incubator is not incidental. The four domains of the call, particularly bio-sourced feedstock resilience and renewable carbon into valuable molecules, map directly onto the platform’s capabilities. For cosmetics, the same model can predict microbial resistance to preservatives and bio-based active ingredients, which is a formulation stability problem that costs the industry time and materials. The incubator’s location at Genopole provides access to synthetic biology and microbiology labs that can validate predictions experimentally, which is the missing piece in my current validation pipeline. The one-year incubation term matches my roadmap from fine-tuning to pilot completion.

I am not claiming product-market fit or revenue; the venture is pre-seed and I am the sole founder. What I am claiming is a validated model with published benchmark performance, a clear technical roadmap with numeric targets, and a named industrial partner for the pilot phase. That is the basis on which I am applying.

ESSAY: SUSTAINABILITY IMPACT

The measurable sustainability impact of this venture is the reduction of experimental waste in drug discovery. Each resistance prediction made in silico replaces a wet-lab experiment that would otherwise consume reagents, plasticware, solvents, and energy. The scale is significant: a single antimicrobial screening campaign can involve tens of thousands of assays, each generating chemical waste and requiring temperature-controlled storage and disposal. My model’s 100 percent mutation coverage means that no mutation is skipped due to missing structural data, so the substitution of computation for experiment is complete rather than partial.

The second impact pathway is in the design of bio-based therapeutics. Molecules synthesized from renewable carbon feedstocks are only sustainable if they reach the market; a candidate abandoned due to resistance is wasted embodied carbon. By predicting resistance early, my platform ensures that bio-based candidates are not discarded for reasons that could have been identified in silico. This is renewable carbon stewardship applied to molecular design.

The third pathway is the transfer to cosmetics. Preservative systems and bio-based active ingredients face the same microbial adaptation pressures as drugs. Predicting resistance to these compounds reduces formulation trial-and-error, which is a source of material waste in product development. The same ESM-2 architecture applies; only the training data changes. This is a direct contribution to L’Oréal’s green sciences mission of sustainable formulation.

ESSAY: TEAM AND COMMITMENT

I am Eniola Olutogun, a pharmacist by training and a machine learning engineer by practice. I built the current model alone, from data curation to benchmark evaluation, and I am the sole author of the venture. This is a deliberate choice at the pre-seed stage: it keeps the technical direction coherent and the equity structure clean. The commitment to the incubator is full-time for the one-year term at Genopole, with relocation to the Île-de-France region already planned.

The team capability is demonstrated by the model’s benchmark results, which required fluency in both protein biology and modern machine learning. The pharmacist background provides the domain knowledge to interpret mutation effects in clinical context; the ML engineering background provides the ability to build and deploy the model. The gap is in industrial partnerships and wet-lab validation, which is precisely what this incubator provides through Genopole’s facilities and L’Oréal’s network. I am not applying for validation of the science; I am applying for the infrastructure to move from benchmark to pilot.

CHECKLIST

- [ ] Confirm application deadline on the L’Oréal Green Sciences Incubator Startup Call 2026 page at hellotomorrowstartupchallenge.submittable.com
- [ ] Verify eligibility for non-incorporated ventures; confirm whether incorporation is required before or during incubation
- [ ] Prepare pitch deck with benchmark results (AUROC 0.804 Platinum, 0.634 SKEMPI 2.0) and roadmap to SKEMPI 2.0 AUROC 0.70
- [ ] Draft one-page technical summary of ESM-2 delta-embeddings plus ECFP4 method for non-specialist reviewers
- [ ] Secure letter of intent from Servier (Suresnes) confirming pilot interest
- [ ] Prepare budget estimate for one-year incubation at Genopole, including compute costs for ESM-2 fine-tuning
- [ ] Confirm whether the call requires a TRL assessment; document current TRL based on validated proof-of-concept
- [ ] Prepare statement on cosmetics transferability: preservative resistance prediction use case
- [ ] Confirm relocation logistics for on-site incubation at Genopole (Evry-Courcouronnes)
- [ ] Verify whether the call requires a legal entity; if so, initiate incorporation in France or EU jurisdiction

EDITOR NOTES

- Eligibility risk: the venture is not yet incorporated; the call may require a legal entity for incubation. Confirm before submitting.
- The SKEMPI 2.0 AUROC of 0.634 is below the stated target of 0.70; the application must present this honestly as the current ceiling and the fine-tuning roadmap as the fix, not claim performance that does not exist yet.
- The cosmetics transferability claim is a bridge, not a current product line; do not overstate it as an active workstream. It is a retraining use case, not a validated result.
- The Servier pilot is named but not contracted; the letter of intent is a required attachment and must be secured before submission.
- The founder is sole author and sole founder; the application should not imply a team exists. The team section must reflect individual capability and the specific gap that Genopole fills.

MOTIVATION LETTER

The L’Oréal Green Sciences Incubator Startup Call 2026 targets ventures that convert renewable carbon into valuable molecules with measurable environmental benefit. My venture does exactly that, but with a computational twist that eliminates the most resource-intensive step in molecular design: experimental resistance screening. I am Eniola Olutogun, a pharmacist and machine learning engineer, and I have built a platform that predicts drug resistance mutations from protein sequence alone, without requiring crystal structures. The model achieves an AUROC of 0.804 on the Platinum benchmark across 553 mutations, with protein-grouped cross-validation, and covers 100 percent of mutations tested versus roughly 18 percent for structure-dependent tools. The proof-of-concept is validated and reproducible.

The green sciences connection is direct. Drug discovery today burns enormous resources on high-throughput experimental screening to identify which candidate molecules will fail due to resistance. Each failed candidate represents embodied carbon, chemical waste, and energy spent on synthesis and assay. My platform predicts resistance in silico, so researchers can discard weak candidates before they reach the lab. For antimicrobial and antiviral programs, this means fewer synthesis cycles, less solvent waste, and a shorter path to molecules that persist in efficacy. That is renewable carbon stewardship applied to pharmaceutical R&D.

I am applying to this specific incubator because Genopole’s facilities and L’Oréal’s innovation network offer what a pre-seed computational venture needs most: wet-lab partners to validate predictions and an industrial lens on bio-based molecule performance. The four domains of this call, particularly renewable carbon into valuable molecules, match my roadmap. I plan to fine-tune the ESM-2 protein language model on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark, then run a pilot with Servier in Suresnes to demonstrate utility in an industrial resistance workflow. The incubator’s one-year on-site term at Genopole aligns with that timeline.

I also see a concrete bridge to cosmetics. Preservative efficacy and microbial resistance to bio-based actives are real problems in formulation science. My model can predict whether microbes will evolve resistance to antimicrobial peptides or preservative systems, which is directly relevant to L’Oréal’s green sciences agenda. The same architecture that predicts drug resistance mutations can be retrained on preservative and active ingredient datasets. This is the same underlying biology of microbial adaptation applied to a different molecular class.

The venture is pre-seed, not yet incorporated, and I am the sole founder. The model is built and benchmarked. What I need from this incubator is the industrial network, the wet-lab validation capacity, and the structured support to move from proof-of-concept to pilot revenue. I am asking for a seat in the 2026 cohort to do that work at Genopole.

RESEARCH STATEMENT

The scientific problem is precise: drug resistance emerges through specific mutations in protein targets, and current prediction tools require crystal structures that exist for only a fraction of clinically relevant proteins. My venture removes that requirement. The method uses ESM-2 protein language model delta-embeddings, which capture the evolutionary and structural context of a mutation from sequence alone, combined with ECFP4 drug fingerprints to represent the ligand. A Random Forest classifier then predicts whether a given mutation confers resistance to a given drug. This architecture is deliberately simple, interpretable, and fast enough to screen entire proteomes in hours.

Validation status is honest and specific. On the Platinum benchmark, which contains 553 mutations across diverse protein-drug pairs, 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 reflects real-world deployment. On SKEMPI 2.0, a harder benchmark with more sparse mutation data, the AUROC is 0.634. That number is the current ceiling and the target for improvement. Published state-of-the-art tools such as mCSM-lig achieve roughly 0.70 AUROC on similar tasks, but they require crystal structures. My model matches or exceeds that performance while covering 100 percent of mutations in the benchmark, versus approximately 18 percent for structure-limited tools. Coverage is the difference between a tool that works in research and a tool that works in production.

The roadmap has three stages. First, fine-tune ESM-2 on the SKEMPI 3K mutation dataset, which contains roughly 3,000 mutation-effect measurements. The goal is to raise SKEMPI 2.0 AUROC from 0.634 to at least 0.70. This is an engineering task with a clear success metric. Second, run a pilot with Servier in Suresnes to apply the model to an active drug discovery program, specifically to pre-screen candidate molecules for resistance liability before experimental validation. The pilot has a defined deliverable: a ranked list of resistance-prone mutations for a Servier target, with experimental confirmation of at least the top five predictions. Third, convert the pilot into a paid subscription or per-project license, establishing the venture’s first recurring revenue.

The green sciences relevance is embedded in the method itself. Every mutation prediction that is correct in silico is an experiment not run in the lab. Each avoided experiment saves reagents, solvents, energy, and labor. For a typical antimicrobial discovery program running 10,000 compound screens per year, even a 20 percent reduction in failed candidates translates to measurable reductions in chemical waste and energy consumption. The platform also enables the design of bio-based therapeutics, which are synthesized from renewable feedstocks, by ensuring those molecules are not abandoned early due to uncharacterized resistance risk. This is renewable carbon into valuable molecules, with the computational layer as the efficiency engine.

The fit with L’Oréal’s Green Sciences Incubator is not incidental. The four domains of the call, particularly bio-sourced feedstock resilience and renewable carbon into valuable molecules, map directly onto the platform’s capabilities. For cosmetics, the same model can predict microbial resistance to preservatives and bio-based active ingredients, which is a formulation stability problem that costs the industry time and materials. The incubator’s location at Genopole provides access to synthetic biology and microbiology labs that can validate predictions experimentally, which is the missing piece in my current validation pipeline. The one-year incubation term matches my roadmap from fine-tuning to pilot completion.

I am not claiming product-market fit or revenue; the venture is pre-seed and I am the sole founder. What I am claiming is a validated model with published benchmark performance, a clear technical roadmap with numeric targets, and a named industrial partner for the pilot phase. That is the basis on which I am applying.

ESSAY: SUSTAINABILITY IMPACT

The measurable sustainability impact of this venture is the reduction of experimental waste in drug discovery. Each resistance prediction made in silico replaces a wet-lab experiment that would otherwise consume reagents, plasticware, solvents, and energy. The scale is significant: a single antimicrobial screening campaign can involve tens of thousands of assays, each generating chemical waste and requiring temperature-controlled storage and disposal. My model’s 100 percent mutation coverage means that no mutation is skipped due to missing structural data, so the substitution of computation for experiment is complete rather than partial.

The second impact pathway is in the design of bio-based therapeutics. Molecules synthesized from renewable carbon feedstocks are only sustainable if they reach the market; a candidate abandoned due to resistance is wasted embodied carbon. By predicting resistance early, my platform ensures that bio-based candidates are not discarded for reasons that could have been identified in silico. This is renewable carbon stewardship applied to molecular design.

The third pathway is the transfer to cosmetics. Preservative systems and bio-based active ingredients face the same microbial adaptation pressures as drugs. Predicting resistance to these compounds reduces formulation trial-and-error, which is a source of material waste in product development. The same ESM-2 architecture applies; only the training data changes. This is a direct contribution to L’Oréal’s green sciences mission of sustainable formulation.

ESSAY: TEAM AND COMMITMENT

I am Eniola Olutogun, a pharmacist by training and a machine learning engineer by practice. I built the current model alone, from data curation to benchmark evaluation, and I am the sole author of the venture. This is a deliberate choice at the pre-seed stage: it keeps the technical direction coherent and the equity structure clean. The commitment to the incubator is full-time for the one-year term at Genopole, with relocation to the Île-de-France region already planned.

The team capability is demonstrated by the model’s benchmark results, which required fluency in both protein biology and modern machine learning. The pharmacist background provides the domain knowledge to interpret mutation effects in clinical context; the ML engineering background provides the ability to build and deploy the model. The gap is in industrial partnerships and wet-lab validation, which is precisely what this incubator provides through Genopole’s facilities and L’Oréal’s network. I am not applying for validation of the science; I am applying for the infrastructure to move from benchmark to pilot.

CHECKLIST

- [ ] Confirm application deadline on the L’Oréal Green Sciences Incubator Startup Call 2026 page at hellotomorrowstartupchallenge.submittable.com
- [ ] Verify eligibility for non-incorporated ventures; confirm whether incorporation is required before or during incubation
- [ ] Prepare pitch deck with benchmark results (AUROC 0.804 Platinum, 0.634 SKEMPI 2.0) and roadmap to SKEMPI 2.0 AUROC 0.70
- [ ] Draft one-page technical summary of ESM-2 delta-embeddings plus ECFP4 method for non-specialist reviewers
- [ ] Secure letter of intent from Servier (Suresnes) confirming pilot interest
- [ ] Prepare budget estimate for one-year incubation at Genopole, including compute costs for ESM-2 fine-tuning
- [ ] Confirm whether the call requires a TRL assessment; document current TRL based on validated proof-of-concept
- [ ] Prepare statement on cosmetics transferability: preservative resistance prediction use case
- [ ] Confirm relocation logistics for on-site incubation at Genopole (Evry-Courcouronnes)
- [ ] Verify whether the call requires a legal entity; if so, initiate incorporation in France or EU jurisdiction

EDITOR NOTES

- Eligibility risk: the venture is not yet incorporated; the call may require a legal entity for incubation. Confirm before submitting.
- The SKEMPI 2.0 AUROC of 0.634 is below the stated target of 0.70; the application must present this honestly as the current ceiling and the fine-tuning roadmap as the fix, not claim performance that does not exist yet.
- The cosmetics transferability claim is a bridge, not a current product line; do not overstate it as an active workstream. It is a retraining use case, not a validated result.
- The Servier pilot is named but not contracted; the letter of intent is a required attachment and must be secured before submission.
- The founder is sole author and sole founder; the application should not imply a team exists. The team section must reflect individual capability and the specific gap that Genopole fills.

AI Draft — L’Oréal Green Sciences Incubator Startup Call 2026