Draft — EWOR Ideation Fellowship

MOTIVATION LETTER

The EWOR Ideation Fellowship exists to back founders building generational impact from outside the traditional innovation hubs. I am that founder. I am a pharmacist who taught myself machine learning engineering in Lagos, Nigeria, because the drug discovery tools I used in clinical practice were failing patients. Resistance mutations to antibiotics and cancer therapies emerge faster than structure-based computational methods can predict them. My venture solves this: it 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. No crystal structure required. On the Platinum benchmark of 553 mutations, my method achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation. This beats the published state-of-the-art tool mCSM-lig, which scores approximately 0.70. More critically, my method covers 100 percent of mutations, while structure-limited tools cover only about 18 percent. This is a paradigm shift in how we predict drug resistance, not an incremental improvement.

I am applying to EWOR because you fund founders who operate at the frontier of physics and biology. My approach is biomathematical CADD: applying frontier physics from protein language models to a problem that kills millions annually in oncology, antivirals, and antimicrobial resistance. I have validated the proof-of-concept. I have not yet incorporated. I am building from Lagos with no EU presence. That is exactly the profile you seek. The 0.1 percent acceptance rate from over 35,000 applicants tells me you are ruthless about selecting founders who can change the world. I am ready for that scrutiny.

My roadmap is concrete. I will fine-tune ESM-2 on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark. I have already initiated a meeting with SEMIA and Quest for Health. I have submitted applications to IncubAlliance and AI House. I have identified Servier in Suresnes as a pilot partner, with a path to annual recurring revenue. The EWOR investment of up to 500,000 euros, including the 110,000 euro fellowship and 390,000 euro convertible note, will fund the compute and wet-lab validation needed to secure that first pharma deal.

I do not need to be in Berlin to build a generational company. I need a partner who sees that the best science often comes from the most unlikely places. EWOR is that partner.

TECHNICAL DEPTH STATEMENT

My venture predicts drug resistance mutations from protein sequence alone. The core innovation is the use of ESM-2 protein language model delta-embeddings. ESM-2 is a transformer model trained on 250 million protein sequences. By computing the difference in embedding vectors between a wild-type and a mutant protein, I capture the biophysical perturbation caused by a single amino acid change. This delta-embedding is a numerical representation of how the protein's internal language changes. I concatenate this with an ECFP4 fingerprint of the drug molecule. The combined feature vector is fed into a Random Forest classifier.

The benchmark results are specific. On the Platinum dataset, which contains 553 mutations across 24 protein targets, my method achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation. This means the model generalizes to unseen proteins, not just unseen mutations. The published state-of-the-art tool mCSM-lig achieves approximately 0.70 on the same benchmark. On SKEMPI 2.0, a more challenging dataset of binding affinity changes, my current AUROC is 0.634. This is the primary target for improvement. By fine-tuning ESM-2 on the SKEMPI 3K mutation set, I expect to push this above 0.70.

The critical advantage is coverage. Structure-based tools require a crystal structure of the protein-drug complex. For most clinically relevant mutations, no such structure exists. My method requires only the protein sequence, which is universally available. This means I can predict resistance for any mutation, including those in emerging viral variants or poorly characterized cancer targets. The 100 percent coverage versus 18 percent for structure-limited tools is a mathematical consequence of the method, not a marketing claim.

MARKET INSIGHT STATEMENT

The global drug discovery market is valued at over 70 billion dollars. The specific bottleneck I address is the failure rate of drugs due to acquired resistance. In oncology, over 90 percent of deaths from metastatic cancers involve resistance to at least one drug. In antimicrobial resistance, the WHO estimates 10 million deaths annually by 2050 if no action is taken. Current computational tools for predicting resistance require crystal structures, which are expensive and time-consuming to obtain. This creates a gap: most resistance mutations are never predicted until they appear in the clinic.

My venture targets three revenue streams. First, direct licensing to large pharma companies for internal drug discovery pipelines. I have identified Servier in Suresnes and Sanofi in Gentilly as named partners. Second, a software-as-a-service platform for biotech startups that lack structural biology capabilities. Third, a data licensing model where the predicted resistance profiles for specific drug-target pairs are sold to research institutions.

The competitive landscape is fragmented. Tools like mCSM-lig, Rosetta, and FoldX all require structural input. My method is the only one that works from sequence alone at comparable or better accuracy. The barrier to entry is the expertise required to train protein language models, which I have developed over three years of self-directed learning. The market is ready for a tool that democratizes resistance prediction. I will capture it.

FOUNDER MOTIVATION STATEMENT

I was a pharmacist in Lagos. I dispensed antibiotics and cancer drugs. I watched patients return with infections that no longer responded to treatment. The resistance was invisible until it was too late. I asked my professors why we could not predict which mutations would cause resistance before the patient took the drug. The answer was always the same: we need a crystal structure, and we do not have one. That answer was unacceptable.

I taught myself machine learning engineering from online courses, papers, and open-source code. I spent two years learning the mathematics of transformers and protein embeddings. I built the first version of my model on a single GPU rented by the hour. The results on the Platinum benchmark were better than any published tool. I knew I had something real.

I am applying to EWOR because you fund founders who have done the hard thing from a hard place. I did not have a lab. I did not have a co-founder. I did not have a network in Berlin or Paris. I had a problem that killed people and a method that could solve it. The EWOR Ideation Fellowship gives me the capital, the mentorship, and the credibility to turn this into a company. The 500,000 euro investment will fund the compute for fine-tuning, the wet-lab validation, and the legal costs of incorporation in France. I will use the virtual mentorship to navigate the European pharma landscape. I will attend the in-person retreats to meet the other founders who are also building generational impact from outside the system.

I am not applying because I want to be a founder. I am applying because I have to be. The patients in Lagos and everywhere else cannot wait for a crystal structure.

CHECKLIST

- [ ] Complete online application at ewor.com/apply
- [ ] Upload motivation letter (this document)
- [ ] Upload technical depth statement (this document)
- [ ] Upload market insight statement (this document)
- [ ] Upload founder motivation statement (this document)
- [ ] Prepare CV/resume highlighting pharmacy degree, ML engineering experience, and benchmark results
- [ ] Prepare one-page pitch deck with AUROC results, coverage comparison, and roadmap
- [ ] Confirm availability for virtual interview
- [ ] Verify no EU presence requirement is still current on EWOR website
- [ ] Prepare list of three references (academic advisor, pharma contact, ML mentor)

EDITOR NOTES

- Eligibility risk: Confirm that EWOR has not changed its global eligibility policy since the research was conducted. The profile states no EU requirement, but verify on the application page.
- Fact verification: The AUROC of 0.804 on Platinum benchmark and 0.634 on SKEMPI 2.0 must be verified against the applicant's actual results. These numbers are specific and will be checked.
- Named partners: The relationships with Servier, Sanofi, and Paris-Saclay institutions need to be confirmed as active conversations, not just identified targets. The application should reflect the actual status.
- Personal detail gap: The applicant should insert a specific anecdote from their pharmacy practice in Lagos that illustrates the problem. This document uses a general statement; a concrete patient story would strengthen the motivation letter.
- Incorporation timeline: The roadmap mentions incorporation in France. The applicant should specify the planned legal structure and timeline, as EWOR may ask about jurisdiction.

MOTIVATION LETTER

The EWOR Ideation Fellowship exists to back founders building generational impact from outside the traditional innovation hubs. I am that founder. I am a pharmacist who taught myself machine learning engineering in Lagos, Nigeria, because the drug discovery tools I used in clinical practice were failing patients. Resistance mutations to antibiotics and cancer therapies emerge faster than structure-based computational methods can predict them. My venture solves this: it 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. No crystal structure required. On the Platinum benchmark of 553 mutations, my method achieves an AUROC of 0.804 plus or minus 0.025 under protein-grouped cross-validation. This beats the published state-of-the-art tool mCSM-lig, which scores approximately 0.70. More critically, my method covers 100 percent of mutations, while structure-limited tools cover only about 18 percent. This is a paradigm shift in how we predict drug resistance, not an incremental improvement.

I am applying to EWOR because you fund founders who operate at the frontier of physics and biology. My approach is biomathematical CADD: applying frontier physics from protein language models to a problem that kills millions annually in oncology, antivirals, and antimicrobial resistance. I have validated the proof-of-concept. I have not yet incorporated. I am building from Lagos with no EU presence. That is exactly the profile you seek. The 0.1 percent acceptance rate from over 35,000 applicants tells me you are ruthless about selecting founders who can change the world. I am ready for that scrutiny.

My roadmap is concrete. I will fine-tune ESM-2 on the SKEMPI 3K mutation dataset to push AUROC above 0.70 on that benchmark. I have already initiated a meeting with SEMIA and Quest for Health. I have submitted applications to IncubAlliance and AI House. I have identified Servier in Suresnes as a pilot partner, with a path to annual recurring revenue. The EWOR investment of up to 500,000 euros, including the 110,000 euro fellowship and 390,000 euro convertible note, will fund the compute and wet-lab validation needed to secure that first pharma deal.

I do not need to be in Berlin to build a generational company. I need a partner who sees that the best science often comes from the most unlikely places. EWOR is that partner.

TECHNICAL DEPTH STATEMENT

My venture predicts drug resistance mutations from protein sequence alone. The core innovation is the use of ESM-2 protein language model delta-embeddings. ESM-2 is a transformer model trained on 250 million protein sequences. By computing the difference in embedding vectors between a wild-type and a mutant protein, I capture the biophysical perturbation caused by a single amino acid change. This delta-embedding is a numerical representation of how the protein's internal language changes. I concatenate this with an ECFP4 fingerprint of the drug molecule. The combined feature vector is fed into a Random Forest classifier.

The benchmark results are specific. On the Platinum dataset, which contains 553 mutations across 24 protein targets, my method achieves an AUROC of 0.804 with a standard deviation of 0.025 under protein-grouped cross-validation. This means the model generalizes to unseen proteins, not just unseen mutations. The published state-of-the-art tool mCSM-lig achieves approximately 0.70 on the same benchmark. On SKEMPI 2.0, a more challenging dataset of binding affinity changes, my current AUROC is 0.634. This is the primary target for improvement. By fine-tuning ESM-2 on the SKEMPI 3K mutation set, I expect to push this above 0.70.

The critical advantage is coverage. Structure-based tools require a crystal structure of the protein-drug complex. For most clinically relevant mutations, no such structure exists. My method requires only the protein sequence, which is universally available. This means I can predict resistance for any mutation, including those in emerging viral variants or poorly characterized cancer targets. The 100 percent coverage versus 18 percent for structure-limited tools is a mathematical consequence of the method, not a marketing claim.

MARKET INSIGHT STATEMENT

The global drug discovery market is valued at over 70 billion dollars. The specific bottleneck I address is the failure rate of drugs due to acquired resistance. In oncology, over 90 percent of deaths from metastatic cancers involve resistance to at least one drug. In antimicrobial resistance, the WHO estimates 10 million deaths annually by 2050 if no action is taken. Current computational tools for predicting resistance require crystal structures, which are expensive and time-consuming to obtain. This creates a gap: most resistance mutations are never predicted until they appear in the clinic.

My venture targets three revenue streams. First, direct licensing to large pharma companies for internal drug discovery pipelines. I have identified Servier in Suresnes and Sanofi in Gentilly as named partners. Second, a software-as-a-service platform for biotech startups that lack structural biology capabilities. Third, a data licensing model where the predicted resistance profiles for specific drug-target pairs are sold to research institutions.

The competitive landscape is fragmented. Tools like mCSM-lig, Rosetta, and FoldX all require structural input. My method is the only one that works from sequence alone at comparable or better accuracy. The barrier to entry is the expertise required to train protein language models, which I have developed over three years of self-directed learning. The market is ready for a tool that democratizes resistance prediction. I will capture it.

FOUNDER MOTIVATION STATEMENT

I was a pharmacist in Lagos. I dispensed antibiotics and cancer drugs. I watched patients return with infections that no longer responded to treatment. The resistance was invisible until it was too late. I asked my professors why we could not predict which mutations would cause resistance before the patient took the drug. The answer was always the same: we need a crystal structure, and we do not have one. That answer was unacceptable.

I taught myself machine learning engineering from online courses, papers, and open-source code. I spent two years learning the mathematics of transformers and protein embeddings. I built the first version of my model on a single GPU rented by the hour. The results on the Platinum benchmark were better than any published tool. I knew I had something real.

I am applying to EWOR because you fund founders who have done the hard thing from a hard place. I did not have a lab. I did not have a co-founder. I did not have a network in Berlin or Paris. I had a problem that killed people and a method that could solve it. The EWOR Ideation Fellowship gives me the capital, the mentorship, and the credibility to turn this into a company. The 500,000 euro investment will fund the compute for fine-tuning, the wet-lab validation, and the legal costs of incorporation in France. I will use the virtual mentorship to navigate the European pharma landscape. I will attend the in-person retreats to meet the other founders who are also building generational impact from outside the system.

I am not applying because I want to be a founder. I am applying because I have to be. The patients in Lagos and everywhere else cannot wait for a crystal structure.

CHECKLIST

- [ ] Complete online application at ewor.com/apply
- [ ] Upload motivation letter (this document)
- [ ] Upload technical depth statement (this document)
- [ ] Upload market insight statement (this document)
- [ ] Upload founder motivation statement (this document)
- [ ] Prepare CV/resume highlighting pharmacy degree, ML engineering experience, and benchmark results
- [ ] Prepare one-page pitch deck with AUROC results, coverage comparison, and roadmap
- [ ] Confirm availability for virtual interview
- [ ] Verify no EU presence requirement is still current on EWOR website
- [ ] Prepare list of three references (academic advisor, pharma contact, ML mentor)

EDITOR NOTES

- Eligibility risk: Confirm that EWOR has not changed its global eligibility policy since the research was conducted. The profile states no EU requirement, but verify on the application page.
- Fact verification: The AUROC of 0.804 on Platinum benchmark and 0.634 on SKEMPI 2.0 must be verified against the applicant's actual results. These numbers are specific and will be checked.
- Named partners: The relationships with Servier, Sanofi, and Paris-Saclay institutions need to be confirmed as active conversations, not just identified targets. The application should reflect the actual status.
- Personal detail gap: The applicant should insert a specific anecdote from their pharmacy practice in Lagos that illustrates the problem. This document uses a general statement; a concrete patient story would strengthen the motivation letter.
- Incorporation timeline: The roadmap mentions incorporation in France. The applicant should specify the planned legal structure and timeline, as EWOR may ask about jurisdiction.

AI Draft — EWOR Ideation Fellowship