Multi-Modal Foundation · tattabio
FlashPPI
A dense retrieval model for rapid, proteome-wide microbial protein-protein interaction screening.
What FlashPPI is
FlashPPI is a computational model developed by TattaBio for predicting protein-protein interactions (PPIs) across entire proteomes. Traditional screening methods require evaluating every candidate pair individually, creating an all-versus-all quadratic computational bottleneck that can take days or months to complete for full proteomes.
By reframing interaction prediction as a dense retrieval task, the system carries out proteome-scale interaction searches in linear time, completing screenings in minutes. It is primarily targeted at microbial proteomes and produces both candidate interaction pairings and residue-level resolution for identified interactions.
How it works
FlashPPI is trained contrastively to perform dense retrieval of interacting protein pairs. To integrate cross-protein and multi-gene co-evolutionary patterns into its representations, the model incorporates genomic priors from the gLM2 language model.
For interaction pairs identified during retrieval, the model evaluates residue-level interactions and predicts fine-grained, two-dimensional residue contact maps to highlight putative binding interfaces. It can also be deployed as a feature-extraction model yielding mean-pooled vector embeddings for input protein sequences.
Key facts
- Task
- Protein-protein interaction prediction and feature extraction
- Input modality
- Protein sequence, text
- Output modality
- Dense embeddings, interaction pairs, and 2D residue contact maps
- Genomic prior source
- gLM2 (650M)
- GPU memory requirement
- 16 GB
- Publication year
- 2026
What you can use it for
- High-throughput proteome screening
- Rapidly screen full microbial proteomes in minutes to identify interacting partner candidates without quadratic computational scaling.
- Residue contact map prediction
- Predict fine-grained 2D residue-residue contact maps to inspect interaction interfaces between retrieved candidate proteins.
- Protein sequence feature extraction
- Generate mean-pooled representation vectors from individual protein amino acid sequences for downstream analysis.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
FlashPPI
Strengths and limits
Good at
- Operates in linear time by formulating interaction prediction as a dense retrieval problem rather than an all-against-all pairwise screening.
- Generates interpretable, residue-level 2D contact maps for retrieved interaction pairs.
- Incorporates genomic priors using gLM2 to capture multi-gene and cross-protein co-evolutionary signals.
Watch out for
- Designed and benchmarked primarily for microbial protein interaction networks.
- Running feature extraction pipelines on typical hardware environments requires at least 16 GB of GPU memory.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use FlashPPI commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run FlashPPI without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
How does FlashPPI achieve linear-time screening?
It casts protein-protein interaction discovery as a dense retrieval task instead of computing every possible protein pair in an exhaustive all-versus-all structural comparison.
Does FlashPPI provide structural details about interactions?
Yes, for candidate pairs identified during retrieval, it predicts fine-grained two-dimensional contact maps at the residue level.
What inputs does the model accept?
It accepts raw protein sequences or FASTA inputs, returning mean-pooled vector embeddings, candidate interactions, and predicted contact maps.
How to cite it
Andre Cornman, Matt Tranzillo, Nicolo G. Zulaybar, Imane Bouzit, and Yunha Hwang. Linear-time prediction of proteome-scale microbial protein interactions. Proceedings of the National Academy of Sciences, 123 (25): e2610619123, 2026. doi:10.1073/pnas.2610619123.
Written from FlashPPI's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.