Predicting human clearance with high accuracy from in silico-derived parameters alone is highly desirable, as it is fast, saves in vitro resources, and is animal-sparing. Random forest (RF) models were generated based on 1340 compounds with human intravenous pharmacokinetic (PK) data, the largest data set publicly available today. The talk highlights ways to assess the general applicability of the developed RF models and shows the difficulty of obtaining useful, broadly applicable models for human clearance. The use of a conformal prediction formalism to derive confidence intervals and assess model applicability is demonstrated. Renally cleared compounds could be identified through an in silico classification model and were found to be better predicted than compounds cleared through other mechanisms. It is hypothesized that this is because renal clearance is mechanistically less complex than e.g. metabolic clearance. The developed RF models compared similar to models derived from scaling human hepatocytes or preclinical in vivo data. The results were generated with the help of KNIME workflows that included MOE nodes, e.g. for descriptor calculations.

Antibody-drug conjugates (ADCs) constitute a rising class of medicines that combine the selectivity of antibodies with the cytotoxicity of drug payloads to yield highly targeted and potent therapeutics. Owing to the need to chemically modify antibody residues for the attachment of the drug payload and their more complex structure compared to either component alone, ADCs can present additional challenges related to stability of the final drug product. This talk explores the use of high-throughput experimental screens and computational techniques to understand the conformational and colloidal behavior of an IgG1-based ADC, representing the first reported example of tuning the biophysical properties (e.g. melting temperature, aggregation temperature, diffusion interaction parameter) of a monomethyl auristatin F-based ADC by screening a large formulation library. The ADC, which exhibits high opalescence and strong attractive protein-protein interactions, is transformed into a more stable structure by experimentally traversing a library of more than ~100 different formulation conditions. A significant reduction in turbidity and diffusion interaction parameter is observed by varying solution attributes such as pH and ionic strength. Molecular modeling simulations rationalized these changes and pointed to the presence of attractive electrostatic interactions between ADC molecules under specific solution conditions (low pH and low ionic strength) facilitated by the drug payload and histidine residues. Experimental characterization of the ADC’s structure using techniques such as circular dichroism, size-exclusion chromatography, and peptide mapping further corroborated the computational results. Taken together, we envision that the studies and findings presented here will be applicable across a wide scope of ADCs towards enhancing their stability, therapeutic efficacy, and ultimate prospects for developability.

Structural information of drug targets has proven to be an asset to drug discovery projects. MOE provides a broad platform for disseminating structural information and project data enabling transforming the data into knowledge to move the project forward. Understanding and quantifying protein-ligand interactions is a valuable consideration in molecular design. Assessing the electrostatic match between compounds and binding pockets provides important insights into why compounds bind. In this presentation I will discuss a case study, where we use MOE project and electrostatic complementary at various stages of a project in medicinal chemistry.