Alignment

Alignment collects notes on comparing proteins by either sequence-derived representations or structural information. The main split here is between sequence-based alignment approaches, including PLM embedding methods and homology retrieval, and structure-based alignment approaches that operate on coordinates, structural descriptors, or structure-derived embeddings.

Quartz 4

Explorer

Alignment

Sequence based

CDR representations segregate into distinct clusters

Distance between averaged PLM embeddings does not correlate with structural difference

Encoder-decoder PLM embeddings are better able to distinguish nearest neighbors than encoder-only PLMs

HMMs cannot identify remote homologs

Larger PLMs are better at homolog detection

PLM embeddings contain enough information to align proteins without fine-tuning

PLM-based sequence searches outperform sequence- and matches structure-based search methods

PLM-derived antibody representations can distinguish engineered from human-derived Abs

PLMs cluster antibodies from the same repertoire by V gene and maturation status

Protein language model embeddings are more predictive of homology than catalytic efficiency

Structural classifications can be learned from PLM-based alignment prediction

Structure-derived embeddings outperform sequence-derived embeddings on LM-based alignment

Template CDR retrieval using embeddings

Contrastive learning of PLM embeddings on functional annotation improves variant effect prediction and homolog detection

Structure based

Discrete structure tokens are worse than continuous representations for structural alignment tasks

Most viral proteins are structurally distinct from eukaryotic proteins

PLM-based sequence searches outperform sequence- and matches structure-based search methods

Protein structure more conserved than sequence

Structure-derived embeddings outperform sequence-derived embeddings on LM-based alignment

Structural models can be aligned at their ligand-binding sites using weighted-exponential term