Video summary

Class 10: ADMET Profiling & Virtual Screening | Bioinformatics Internship 2026

Main summary

Key takeaways

Science and Nature

Scientific concepts / discoveries / nature phenomena covered

  • ADMET profiling (ADMET = Absorption, Distribution, Metabolism, Excretion, Toxicity) Used as a drug-discovery filter before molecular docking and wet-lab testing.

  • Pharmacokinetics (PK): what the body does to the drug Expanded into ADMET when combined with toxicity. Contrasted with pharmacodynamics (PD): what the drug does to the body.

Core PK sub-processes and key factors

Absorption

  • Oral bioavailability / fraction absorbed
  • Permeability of the intestinal epithelium (e.g., Caco-2 cell model)
  • P-glycoprotein (P-gp) efflux
    • Substrates/inhibitors affecting intestinal transport
  • First-pass metabolism by the liver before systemic circulation

Distribution

  • Volume of distribution (Vd): distribution across blood vs tissues
  • Plasma protein binding (PPB)
    • Example: albumin binding
    • Focus on the free drug fraction
  • Blood–brain barrier (BBB) penetration and CNS selectivity

Metabolism

  • Cytochrome P450 enzymes (explicitly mentioned)
    • CYP3A4, CYP2D6, CYP2C9 (also generally the CYP family)
  • Substrate vs inhibitor vs inducer behavior
    • Drives drug–drug interaction (DDI) risk
  • Typical reaction types referenced:
    • oxidation/reduction, hydrolysis, conjugation pathways
  • Active metabolites concept
    • Metabolite potency vs parent compound

Excretion

  • Renal elimination
    • glomerular filtration and tubular secretion
  • Fecal/biliary elimination (mentioned)
  • Half-life
    • Time for plasma concentration to fall to ~50%
  • Clearance
    • Plasma cleared per unit time

Toxicity prediction concepts

  • Cardiac toxicity via HERG / Kv channel (hERG) blockage
    • Leads to risk of QT prolongation
  • Genotoxicity / mutagenicity
    • Conceptually tied to Ames test
  • Drug-induced liver injury (DILI) mechanisms (mentioned)
    • reactive metabolites
    • mitochondrial toxicity
    • immune-mediated toxicity
  • Other toxicity categories referenced
    • carcinogenicity
    • reproductive toxicity
    • skin sensitization
    • emetic/other risks (as stated)
  • Structural alerts / substructures associated with toxicity
    • e.g., nitro groups, aromatic amines, epoxides, alkylating agents
  • Threshold-style risk interpretation for QT risk
    • Ranges based on IC50 / micromolar values (as described)

Rule-based physicochemical filtering in ADMET workflows

  • Lipinski’s Rule of Five
  • Rule of Three
  • Physicochemical inputs/criteria mentioned, such as:
    • molecular weight
    • logP
    • H-bond donors/acceptors
    • polar surface area (PSA)
    • rotatable bonds, etc.

Virtual screening pipeline themes (compound set → docking-ready library)

  • Transform raw ligand libraries into docking-ready 3D structures
    • after ADMET/toxicity filtering
  • Reduce redundancy using structural diversity handling
    • similarity measures and clustering (e.g., “St. Motif similarity”)

Computational tools and data formats for ligand preparation

  • Cameo / CAMBRL
    • retrieving bioactivity data linked to targets (e.g., IC50, EC50, inhibition constants, dissociation constants, etc.)
  • Chemical file formats:
    • SMILES (1D representation)
    • SDF (structure data file)
    • PDBQT (protein docking ligand format)
    • 2D → 3D conversion and subsequent conversion steps for docking readiness
  • RDKit
    • extracting SMILES and calculating properties
  • Open Babel / Open Web(…)/(typoed)
    • format conversion
  • Chimera / “aricut” / Ar… tools (as mentioned)
    • cleaning/canonicalizing SMILES/CSV
    • removing problematic tokens (e.g., salts, empty lines, etc.)

Virtual screening justification

  • Screen safe/filtered candidates first to reduce cost and time before docking and wet-lab trials.
  • Emphasis that later-stage failures are expensive (market development cost referenced).

Methodology / workflow outlined

Target-to-ligand sourcing (bioactivity-based library building)

  1. Start with a target (from prior identification).
  2. Query a bioactivity database for compounds with experimental bioassay results against the target.
  3. Filter/select compounds using criteria such as:
    • clinical phase
    • Lipinski violations (as described)

Download and prepare ligand datasets

  • Download compound data as CSV and SDF
    • SDF used as the structure library
  • Convert/clean data:
    • fix CSV delimiter/formatting (semicolon separation described)
    • extract SMILES into a clean single-column smiles file
    • remove/handle salts and artifacts so prediction tools can ingest inputs

ADMET / toxicity prediction (in silico)

  • Workflow order emphasized: toxicity first
    • run toxicity prediction on the prepared structure/SMILES dataset
    • if non-toxic/acceptable, proceed to broader ADMET predictions
      • absorption, distribution, metabolism, excretion
  • Evaluate model outputs including:
    • Caco-2 permeability and thresholds (Papp in cm/s units)
    • Oral bioavailability / oral fraction
      • confidence categories (low/medium/high)
    • BBB penetration scores and criteria
      • example thresholds mentioned: molecular weight ~<450 Da, low pKa range, PSA <~60–90 Ų, etc.
    • P-gp substrate flags (to reduce CNS penetration)
    • PPB percentage
      • implications for free fraction and half-life/half-lives
    • CYP450 interactions
      • substrate/inhibitor/inducer risk and DDI potential
    • hERG/QT risk categorization

Select final “docking-ready” candidates

  • Keep molecules that are:
    • acceptable in ADMET/toxicity (especially non-toxic)
    • have adequate absorption/distribution metrics for intended route (oral/CNS vs peripheral)
  • Apply diversity/cluster filtering
    • maintain a structurally diverse set

Prepare docking-ready ligand structures

  • Convert:
    • 2D SDF → 3D SDF → PDBQT
  • Use format conversion and ligand preparation tools (e.g., Open Babel and related steps).

Next step

  • Proceed to molecular docking protocols after ADMET profiling.

Researchers / sources featured (named at the end)

Named individuals

  • No specific individual researchers were clearly identified as scientific authors in the content.
  • Featured speaker credited:
    • Miss Adiba Fatima (also appears as “Miss Aba Fatima” in spelling variants)

Named tools / databases / software

  • ChEMBL (spoken as “Camble/Cambal”; implied)
  • DrugBank
  • UniProt (referenced from a previous session)
  • NCBI GeneCards (referenced from a previous session)
  • PDB (Protein Data Bank)
  • RDKit
  • Open Babel / Open Web(….) (name appears as “open webble” in the text)
  • Deep PK / DPK (deep learning ADMET prediction tool)
  • SwissADME (mentioned as unavailable in practice)
  • ADMETlab
  • Ames test (standard assay)
  • Lipinski’s Rule of Five (standard medicinal chemistry guideline)

Original video