Tesorai Refine
Reducing false discovery in high-throughput assays
High-throughput assays are built for speed/scale, at the cost of high false discovery rates (in some cases over 90%)
Proteomics and mass spectrometry data is instrumental in discovery efforts but can be challenging to interpret
Types of assays
Affinity purification with mass spectrometry (APMS), covalent fragment screens with mass spectrometry, DNA-encoded libraries, phage displays, small-molecule binding/activity assays, ribosome profiling
False discovery rate
Assays are noisy, resulting in high false-positive rates (for example, only ~5% of interactions detected in APMS are biologically relevant), with likely large false-negative rates too
Challenges with interpreting data
Biases like sampling and detection favor abundant analytes, alongside unspecific interactions and plate/batch effects, impacting analysis. Limited labels hinder standard AI/ML in custom applications.
Impact
Relying on intensities measured by the assays alone can be misleading, and can lead to high failure rates
Tesorai is replacing current experiment-constrained manual engineering and simplistic modeling with data integration and deep learning
Current ranking model improves upon the gold standard ranking algorithm
Identification of High Potential Protein-Protein Interactions (PPI)
* This evaluation is performed on the ProteomeTools dataset (a set of chemically synthesized peptides)
** This rate represents an identification rate of 92%
Identification of High Potential Protein-Protein Interactions (PPI)
* evaluated on a left-out organism and PRIDE project (Arabidopsis thaliana)
Assay types that Onyx suite can be applied to
Identification of new Protein-Protein Interactions (PPI)
Affinity purification & mass spectrometry (AP-MS)
Antibody-antigen mapping
Molecular glues validation
Identification of New Protein- Ligand Interactions (PLI)
Chemoproteomics
Hit identification from High-throughput assays, including by improving batch and plate effects
DNA-encoded libraries
Improving properties of individual molecules
Protein stability
Peptide solubility
Small-molecule library design
Custom Fine-Tuned AI Model
We can leverage our foundation models to address unique challenges in analyzing high-throughput assays with proteins, peptides, and/or small molecules.
Book a Call
Free consultation on addressing challenges your team is facing and/or advice on optimizing data science efforts.