LambdaPP: Fast and accessible protein-specific phenotype predictions

Abstract

The availability of accurate and fast Artificial Intelligence (AI) solutions predicting aspects of proteins are revolutionizing experimental and computational molecular biology. The webserver LambdaPP aspires to supersede PredictProtein, the first internet server making AI protein predictions available in 1992. Given a protein sequence as input, LambdaPP provides easily accessible visualizations of protein 3D structure, along with predictions at the protein level (GeneOntology, subcellular location), and the residue level (binding to metal ions, small molecules, and nucleotides; conservation; intrinsic disorder; secondary structure; alpha-helical and beta-barrel transmembrane segments; signal-peptides; variant effect) in seconds. The structure prediction provided by LambdaPP - leveraging ColabFold and computed in minutes - is based on MMseqs2 multiple sequence alignments. All other feature prediction methods are based on the pLM ProtT5. Queried by a protein sequence, LambdaPP computes protein and residue predictions almost instantly for various phenotypes, including 3D structure and aspects of protein function.

Ultra-fast protein structure prediction to capture effects of sequence variation in mutation movies

“…Due memory constraints, we removed proteins >430 residues, yielding 93,286 sequences for training. To avoid bias from overrepresented families, we clustered the training set with UniqueProt 50 at the default Hval=0. Following AlphaFold developers 4 , we trained on randomly picked samples from the resulting 11,580 clusters.…”

Alignment-based protein mutational landscape prediction: doing more with less

“…The proteins sharing the same domain composition will have exactly the same MSAs. To avoid such redundancy, we focused on a subset of 59 proteins extracted with an adjusted version of UniqueProt [29, 30]. Instead of PSI-BLAST we used MMseqs2 to improve runtime, and discarded alignments of less than 50 residues for pairs of sequences with at least 180 residues to prevent very short alignments from removing longer sequences.…”

Refining Embedding-Based Binding Predictions by Leveraging AlphaFold2 Structures

“…Calculating the structure bonus values for all three ligand classes of all 1010 proteins and writing them to disk only took around 5 seconds, resulting in a runtime of around 5 milliseconds per protein. We excluded the time for computing the structure and binding predictions from the runtime analysis as with the release of the fourth version of AlphaFold Protein Structure Database (AFDB), 200 million protein structures are available [21] for download and tools like bio-embeddings [23] and LambdaPP [24] allow for fast and easy bindEmbed21DL predictions.…”