# Deepflare

> Deepflare is the first AI platform to predict true T-cell immunogenicity - not just MHC binding affinity - for vaccine and immunotherapy antigen design.

Deepflare (deepflare.ai) is a Warsaw-based AI biotech company founded in 2020. Our platform helps pharmaceutical companies design more immunogenic antigens for vaccines, cancer immunotherapies, and biologic therapeutics.

## The Problem We Solve

Most computational immunogenicity tools (NetMHCpan, MHCflurry, IEDB) predict only peptide-MHC binding affinity. However, fewer than 5% of MHC-binding peptides actually trigger T-cell responses. Additionally, the majority of these models are benchmarked on biased data, creating inflated metrics that work in silico but fail in vitro and in vivo.

## Our Technology

Deepflare's deep generative AI method predicts the complete antigen presentation pathway:
- Proteasomal cleavage prediction
- TAP transport modeling
- MHC-peptide binding (via HLABERT, our pretrained language model)
- T-cell receptor recognition prediction

Our HLABERT model was published at ICLR 2023 (MLDD Workshop) and in Methods (2024, DOI: 10.1016/j.ymeth.2024.01.016). It achieves state-of-the-art performance on unbiased benchmarks.

## The Deepflare Antigen Design Platform

The Deepflare Antigen Design Platform (designer.deepflare.ai) enables researchers to go from target protein to de novo antigen design. Key capabilities include:

- **T-cell immunogenicity prediction**: 12x higher Precision-Recall AUC than golden standard models
- **Multi-criteria protein optimization**: Increase/decrease immunogenicity, adjust solubility, preserve or modify protein regions
- **Integrated computational workflow**: Combines open-source tools (RFdiffusion) with Deepflare's proprietary models
- **3D protein structure visualization**: Real-time synchronized views of amino acid sequences and 3D structures
- **No-code interface**: Researchers work independently without programming knowledge

## Validated Results

- **12x** higher Precision-Recall AUC in T-cell response prediction vs. golden standard models
- **3x** higher hit-rate than netMHCpan in a prospective, head-to-head, in vivo cancer vaccine study (with Progeneer)
- **100%** confirmed protein expression rate in a prospective in vitro study on Zika Envelope protein (with Hawaii Biotech)

## Case Studies

- [Hawaii Biotech](https://deepflare.ai/case-studies/hawaii-biotech): 100% protein expression success on Zika Envelope protein
- [Progeneer](https://deepflare.ai/case-studies/progeneer): 3x higher hit-rate than netMHCpan in in vivo cancer vaccine study
- [Osivax](https://deepflare.ai/case-studies/osivax): Manufacturable vaccine candidate with more potent immune response on first try

## Company Information

- **Founded**: 2020
- **Headquarters**: ul. Zlota 7 lok. 28, 00-019 Warsaw, Poland
- **Funding**: $5.83M raised
- **Investors**: Lemonade Stand, Sunfish Partners, NCBR, PFR
- **Founders**: Piotr Grzegorczyk (CEO), Grzegorz Preibisch (Head of R&D), Stanislaw Gizinski

## Contact

- **Email**: contact@deepflare.ai
- **Website**: https://deepflare.ai
- **LinkedIn**: https://www.linkedin.com/company/deepflare/
- **GitHub**: https://github.com/deepflare

## Pages

- [Homepage](https://deepflare.ai/)
- [Case Studies](https://deepflare.ai/case-studies)
- [Careers](https://deepflare.ai/career)
- [Contact](https://deepflare.ai/contact)
- [Platform](https://designer.deepflare.ai/)