iBioProVis

iBioProVis maps compounds into two-dimensional spaces using t-Distributed Stochastic Neighbor Embedding (t-SNE) and principal component analysis (PCA) to analyze structural similarity and bioactivity relationships between compounds (optionally provided as SMILES) and target proteins using a curated ChEMBL v25 bioactivity dataset.

Key Features:

• ChEMBL v25 dataset: Uses a curated ChEMBL (version 25) dataset encompassing over 15 million bioactivity measurements.
• Filtering and pre-processing: Performs meticulous filtering and pre-processing to generate a reliable compound-target dataset.
• Input types: Accepts target protein identifiers and optional compound SMILES as input.
• Non-linear dimensionality reduction (t-SNE): Applies t-Distributed Stochastic Neighbor Embedding (t-SNE) to project compounds into a two-dimensional space.
• Principal Component Analysis (PCA): Provides a PCA projection alongside t-SNE embeddings for comparative dimensionality reduction.
• Structural and bioactivity mapping: Embeddings reflect structural similarities among compounds and relationships to associated targets, facilitating detection of patterns related to shared binding affinities.
• Database cross-references: Integrates cross-references to established databases to provide contextual information on drugs and drug candidate compounds.
• Hypothesis generation support: Enables inference of potential new binders for proteins or candidate target proteins for compounds by analyzing spatial proximity in embedding space.

Scientific Applications:

• Drug discovery and pharmacology: Exploration of compound-target interaction space to support candidate selection and repurposing hypotheses.
• Bioactivity analysis: Comparative assessment of compound distributions and bioactivity patterns using t-SNE and PCA projections.
• Target binder identification: Generation of hypotheses for potential novel binders based on proximity to known ligands in embedding space.
• Case study — ACE2 and SARS-CoV-2: Analysis of ACE2 binding compounds and antiviral drugs, including compounds considered in clinical trials for COVID-19.

Methodology:

Uses a curated ChEMBL v25 bioactivity dataset (>15 million measurements) with filtering and pre-processing to produce a compound-target dataset; accepts target protein identifiers and optional SMILES; applies t-SNE for non-linear 2D embedding and provides PCA projections, with embeddings reflecting structural similarity and bioactivity relationships.