Augmenting DMTA using predictive AI modelling at AstraZeneca

Identification of Novel Potent NSD2-PWWP1 Ligands Using Structure-Based Design and Computational Approaches

“…In order to have a better understanding of the rat hepatic clearance of 34, an in silico assessment was explored using the AstraZeneca bespoke artificial intelligence (AI) predictive modeling infrastructure. 33 In addition to a predicted numerical value for intrinsic clearance, this model also generates glowing maps which highlight the regions of the molecule more likely to be enzymatically modified (Figure 8). The intrinsic clearance of 34 in rat hepatocytes was predicted as 81 μL/min/1 × 10 −6 , whereas four structural regions were highlighted as the major sites of metabolism (Figure 8): (1) the saturated carbon on the head region, (2) the methyl substituent of the isoxazole central core, (3) the methyl group of the phenyl ring of the tail region, and (4) the solventexposed piperidine ring.…”

“…In view of the favorable profile of compound 34 in the Caco-2 assay (Table ), we ascribed its poor pharmacokinetic behavior to hepatic metabolism rather than a permeability limitation. In order to have a better understanding of the rat hepatic clearance of 34 , an in silico assessment was explored using the AstraZeneca bespoke artificial intelligence (AI) predictive modeling infrastructure . In addition to a predicted numerical value for intrinsic clearance, this model also generates glowing maps which highlight the regions of the molecule more likely to be enzymatically modified (Figure ).…”

“…Pure fractions were evaporated to dryness to afford 3-(4bromo-5-methylisothiazol-3-yl)-4-methylbenzaldehyde (0.4 g, 73%) as a yellow gum. 1 [1,4]oxazin-3(4H)-one (33). Piperidine (234 mg, 2.7 mmol) was added to a stirred solution of 4-methyl-3-(5methyl-4-(3-oxo-3,4-dihydro-2H-benzo[b] [1,4]oxazin-7-yl)isothiazol-3-yl)benzaldehyde (100 mg, 0.3 mmol) in DCM (5 mL) at RT.…”

QSARtuna: An Automated QSAR Modeling Platform for Molecular Property Prediction in Drug Design

“…Other applications include the prediction of chemical reaction yields, where reactants and yield are provided as training data. 19 The development of novel algorithms capable of rationalizing complex relationships between chemical and biological information, 20,21 exponentially growing chemical and biological space added to molecule databases, 22 falling cost of computational resources, 23,24 and MLOps systems for accessing production-level models 25 have spearheaded the development and use of QSAR models in practice. 26−28 Despite this, the assortment of workflows, algorithmic methods, and parameters means training and updating models is nontrivial and finding the relatively optimal modeling setup is a time-consuming task for data scientists.…”

“…The development of novel algorithms capable of rationalizing complex relationships between chemical and biological information, , exponentially growing chemical and biological space added to molecule databases, falling cost of computational resources, , and MLOps systems for accessing production-level models have spearheaded the development and use of QSAR models in practice. − Despite this, the assortment of workflows, algorithmic methods, and parameters means training and updating models is nontrivial and finding the relatively optimal modeling setup is a time-consuming task for data scientists. Consequently, there is a need to compare different models for specific properties reproducibly, efficiently, and robustly across different molecular representations and algorithms.…”