Chemoinformatics in Action

Chemoinformatics in Action

• 1.
  Chemoinformatics in action: some question for audience Yuriy Sushko, Sergii Novotarskyi
• 2.
  Practical example Story: A companythat produces or intends to produce some particular compound (drug, make up, paint, glue, toilet refresher, whatever..) is obliged to test, if this compound is toxic for human and how toxic it is. What are the options to check this? Teuthrin, Cyclopropanecarboxylic acid
• 3.
  Practical example Bioassay Computer modeling In silico: using QSAR (QSPR) based on machine learning to predict In vivo and in vitro assays with properties of interest without direct mice, dogs, rats or other species experiment.
• 4.
  Option 1: Bioassay Classicaland currently widely used method for measuring toxicity is bioassay with mice, rats, dogs or other species. What are advantages and disadvantages
• 5.
  Option 1: Bioassay Forbioassay we would typically need: • Dozens of mice for checking several concentrations of tested compound • In some assays we need to wait for next generation • We may need to test against several organisms (rat, mouse) and dierent administration routes (oral, skin, IV injection) • Test can take upto several months • Test would cost upto dozens of thousands dollars What if we need to measure toxicity for 100 000 compounds?
• 6.
  Option 2: Modeling Whatare the steps required to build predictive model for physicochemical or biological property? • Prepare dataset of experimental data • Choose and calculate molecular descriptors • Apply machine learning method
• 7.
  Molecular descriptors What isdescriptor? Most simple examples? Descriptor is some numerical property of chemical compound. • Simplest constitutional descriptors: MW, NA, nDB, .. • Molecular properties: LogP, hydrophilic factor, .. • Randic molecular profiles • Various topological and 3D indices and profiles
• 8.
  Molecular descriptors 2.54 4.25 -5.71 3.26 0.57 -0.07 1.45 6.34 8.28 2.78 -5.67 -2.33 1.45 7.34 8.35 1.64 -5.56 -4.45
• 9.
  Machine learning What kindof machine learning methods do you know? • Linear regression • K nearest neighbors (KNN) • Partial Least Regression • Neural networks • Support Vector Machines
• 10.
  Some additional facts Popularformats for representing molecules in databases • SDF • SMILES • INCHI
• 11.
  SDF — aplain text file benzene ACD/Labs0812062058 header 6 6 0 0 0 0 0 0 0 0 1 V2000 1.9050 -0.7932 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 1.9050 -2.1232 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 0.7531 -0.1282 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 0.7531 -2.7882 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 atom information -0.3987 -0.7932 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 -0.3987 -2.1232 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0 2 1 1 0 0 0 0 3 1 2 0 0 0 0 4 2 2 0 0 0 0 5 3 1 0 0 0 0 bond information 6 4 1 0 0 0 0 6 5 2 0 0 0 0 M END $$$$ > <Unique_ID> XCA3464366 > <ClogP> 5.825 tags > <Vendor> Sigma > <Molecular Weight> 499.611
• 12.
  SMILES — astring representation C1=CC=C(C=C1)Br CC(F)F COC(C(Cl)Cl)(F) F
• 13.
  InChI — onemore approach InChI (international chemical identifier) — a standart, developed by IUPAC for a textual identifier of chemical substances InChI: InChI=1S/C6H5Br/c7-6-4-2-1-3-5-6/h1-5H InChIKey: QARVLSVVCXYDNA-UHFFFAOYSA InChI: InChI=1S/C2H4F2/c1-2(3)4/h2H,1H3 InChIKey: NPNPZTNLOVBDOC-UHFFFAOYSA InChI: InChI=1S/C3H4Cl2F2O/c1-8-3(6,7)2(4)5/h2H,1H3 InChIKey: RFKMCNOHBTXSMU-UHFFFAOYSA