Using Generative Modeling to Endow with Potency Initially Inert Compounds with Good Bioavailability and Low Toxicity

In the early stages of drug development, large chemical libraries are typically screened to identify compounds of promising potency against the chosen targets. Often, however, the resulting hit compounds tend to have poor drug metabolism and pharmacokinetics (DMPK), with negative developability features that may be difficult to eliminate. Therefore, starting the drug discovery process with a “null library”, compounds that have highly desirable DMPK properties but no potency against the chosen targets, could be advantageous. Here, we explore the opportunities offered by machine learning to realize this strategy in the case of the inhibition of α-synuclein aggregation, a process associated with Parkinson’s disease. We apply MolDQN, a generative machine learning method, to build an inhibitory activity against α-synuclein aggregation into an initial inactive compound with good DMPK properties. Our results illustrate how generative modeling can be used to endow initially inert compounds with desirable developability properties.


Compounds and chemicals
Compounds were purchased from MolPort (Riga, Latvia) or Mcule, and prepared in DMSO to a stock of 5 mM.All chemicals used were purchased at the highest purity available.

Recombinant αS expression
Recombinant αS was purified as described previously [1][2][3] .The plasmid pT7-7 encoding for human αS was transformed into BL21-competent cells.Following transformation, competent cells were grown in LB in the presence of ampicillin (100 μg/mL).Cells were induced with IPTG and grown overnight at 37 o C and harvested by centrifugation in a Beckman Avanti J25 centrifuge with a JA-20 rotor at 5000 rpm (Beckman Coulter, Fullerton, CA).The cell pellet was resuspended in 10 mM Tris, pH 8.0, 1 mM EDTA, 1 mM PMSF and lysed by multiple freeze-thaw cycles and sonication.The cell suspension was boiled for 20 min and centrifuged at 13,500 rpm with a JA-20 rotor (Beckman Coulter).Streptomycin sulfate was added to the supernatant to a final concentration of 10 mg/mL and the mixture was stirred for 15 min at 4 o C.After centrifugation at 13,500 rpm, the supernatant was taken with an addition of 0.36 g/mL ammonium sulfate.The solution was stirred for 30 min at 4 o C and centrifuged again at 13,500 rpm.The pellet was resuspended in 25 mM Tris, pH 7.7, and ion-exchange chromatography was performed using a HQ/M-column of buffer A (25 mM Tris, pH 7.7) and buffer B (25 mM Tris, pH 7.7, 600 mM NaCl).The fractions containing αS (≈ 300 µM) were dialysed overnight against the appropriate buffer.The protein concentration was determined spectrophotometrically using ε280 = 5600 M −1 cm −1 .

Seed fibril preparation
αS fibril seeds were produced as described previously 1,2 .Samples of αS (700 µM) were incubated in 20 mM phosphate buffer (pH 6.5) for 72 h at 40 °C and stirred at 1,500 rpm with a Teflon bar on an RCT Basic Heat Plate (IKA, Staufen, Germany).Fibrils were then diluted to 200 µM, aliquoted and flash frozen in liquid N2, and finally stored at -80 o C. For the use of kinetic experiments, the 200 µM fibril stock was thawed, and sonicated for 15 s using a tip sonicator (Bandelin, Sonopuls HD 2070, Berlin, Germany), using 10% maximum power and a 50% cycle.

Measurement of aggregation kinetics
αS was injected into a Superdex 75 10/300 GL column (GE Healthcare) at a flow rate of 0.5 mL/min and eluted in 20 mM sodium phosphate buffer (pH 4.8) supplemented with 1 mM EDTA.The obtained monomer was diluted in buffer to a desired concentration and supplemented with 50 µM ThT and preformed αS fibril seeds.The molecules (or DMSO alone) were then added at the desired concentration to a final DMSO concentration of 1% (v/v).Samples were prepared in low-binding Eppendorf tubes, and then pipetted into a 96-well half-area, black/clear flat bottom polystyrene NBS microplate (Corning 3881), 150 µL per well.The assay was then initiated by placing the microplate at 37 o C under quiescent conditions in a plate reader (FLUOstar Omega, BMG Labtech, Aylesbury, UK).The ThT fluorescence was measured through the bottom of the plate with a 440 nm excitation filter and a 480 nm emission filter.After centrifugation at 5000 rpm to remove aggregates the monomer concentration was measured via the Pierce™ BCA Protein Assay Kit according to the manufacturer's protocol.

Toxicity measurement
Human neuroblastoma cells (SH-SY5Y) were cultured in DMEM/F-12 GlutaMAX™ (#10565018, Gibco) supplemented with 10% heat inactivated foetal bovine serum (hiFBS) (#10082147, Gibco).Cells were kept at 37C, 5% CO2 and 95% relative humidity, unless otherwise stated.For each experiment, cells were plated in tissue culture-treated, flat bottom, white polystyrene plates (#CLS3917, Corning) at a final density of 10k or 20k cells/well.Cells were incubated for 24h prior to treatment to ensure attachment.On the day of the treatment, drugs were sonicated for 10 minutes and filtered using a 0.02 μm Whatman Anotop filter.100X stocks from each molecule were prepared in a sterile mirror plate (#CLS3997, Corning) by performing log-serial dilutions in DMSO.Intermediate drug stocks were diluted in serum free DMEM/F-12 GlutaMAX™ to a final 10X concentration.Mirror plates were incubated for 10 minutes at 550 rpm.30 minutes prior to the addition of the drugs, culture medium was replaced with serum free DMEM/F-12 GlutaMAX™.90 μl of medium was added per well.10μl of each drug