Informatic challenges and advances in illuminating the druggable proteome

Abstract

The understudied members of the druggable proteomes offer promising prospects for drug discovery efforts. While large-scale initiatives have generated valuable functional information on understudied members of the druggable gene families, translating this information into actionable knowledge for drug discovery requires specialized informatics tools and resources. Here, we review the unique informatics challenges and advances in annotating understudied members of the druggable proteome. We demonstrate the application of statistical evolutionary inference tools, knowledge graph mining approaches, and protein language models in illuminating understudied protein kinases, pseudokinases, and ion channels.

Keywords: machine learning; network biology; orthology; protein evolution; sequence embedding.

FIGURE 1

The panels represent the informatics approaches used for illuminating the druggable proteome. (a) Protein sequence exploration tools that enable analysis based on large collections of sequences, such as detection and collection of orthologs (KinOrtho), motif-based clustering of related sequences (BPPS), and embedding based-analyses (kibby and chumby). (b) Protein ontology exploration tools that rely on visualizing and querying prebuilt knowledge graphs to infer feature associations and predict functional and regulatory roles. (c) A schematic showing how the tools in panels (a) and (b) can be used iteratively to compare and contrast an understudied protein sequence with its well-studied counterparts to derive novel hypotheses and identify drug targets.

FIGURE 2

A demonstration of the ProtVista viewer within the ProKinO browser for mapping sequence annotations to 3D models using the p21 activated kinase 5 (PAK5) as an example. (a) Snapshot of the ProtVista viewer showing key sequence annotations within the ligand binding site of PAK5. (b) Zoomed-in view of the ATP/ligand binding residues. (c) Schematic of a small molecule targeting the active site of PAK5.

FIGURE 3

(a) An embedding tree of the human Cys-loop ligand-gated ion channels using the transmembrane region. Nodes with a black star indicate the dark ion channels in this subfamily. To the left of the tree, we plot a UMAP projection using the same dataset. (b) A condensed tree showing various families of Cys-loop ligand-gated ion channels. The red percentage values indicate VIBE scores. (c) Cryo-electron microscopy structures of an example membrane-bound ligand-gated ion channel, human α1β2γ2 GABA_A receptor.^(p86) The structural model on the left shows the heteropentameric assembly built from three unique subunits: α1, β2, and γ2. Each monomer consists of an N-terminal extracellular domain, which adopts a β-sandwich, followed by the C-terminal transmembrane domain. Within the pentamer, an α1 subunit is colored white. The structure on the right depicts the transmembrane domain of an α1 subunit. (d) Embedding-based sequence conservation for the transmembrane region of GABA_A receptor subunits α1, β2, and γ2. Each consists of four transmembrane helices, designated M1–4, and a disordered segment on the M3–4 loop. Bars indicate the level of conservation at that sequence position, with taller bars indicating higher conservation.