Calculations have been made of transport coefficients of electrons in gas mixtures for ratios CO2:N2:He of 1:1:8, 1:2:3, 1:7:30, and 1:0.25:3. New cross sections for CO2 derived from swarm experiments are used together with previously published cross sections for N2 and He. Curves are presented of the predicted electron drift velocity, transverse and longitudinal diffusion coefficients, and ionization and attachment coefficients for E/N values ranging from 10−18 to 1 × 10−15 V cm2; E is the electric field strength and N the gas number density. Examples are given of derived distribution functions and comparisons are made with a Maxwellian distribution function. The percentage of the input electrical power which excites vibrational processes coupled to the 001 upper laser level of CO2 is given as a function of E/N. The maximum efficiency from these calculations increases for increasing ratios of N2:CO2, because the proportion of energy used to excite the bending and stretching modes of CO2 is then reduced. By assuming a recombination coefficient of 10−7 cm3 sec−1, the operating E/N for self-sustained glow discharges is predicted as a function of current density for various laser mixtures by equating the ionization rate to the attachment and recombination rate.

We describe a novel deep learning neural network method and its application to impute assay pIC 50 values. Unlike conventional machine learning approaches, this method is trained on sparse bioactivity data as input, typical of that found in public and commercial databases, enabling it to learn directly from correlations between activities measured in dierent assays.In two case studies on public domain data sets we show that the neural network method outperforms traditional quantitative structureactivity relationship (QSAR) models and other leading approaches. Furthermore, by focussing on only the most condent predictions the accuracy is increased to R 2 > 0.9 using our method, as compared to R 2 = 0.44 when reporting all predictions.

Contemporary deep learning approaches still struggle to bring a useful improvement in the field of drug discovery because of the challenges of sparse, noisy, and heterogeneous data that are typically encountered in this context. We use a state-of-the-art deep learning method, Alchemite, to impute data from drug discovery projects, including multitarget biochemical activities, phenotypic activities in cell-based assays, and a variety of absorption, distribution, metabolism, and excretion (ADME) endpoints. The resulting model gives excellent predictions for activity and ADME endpoints, offering an average increase in R 2 of 0.22 versus quantitative structure–activity relationship methods. The model accuracy is robust to combining data across uncorrelated endpoints and projects with different chemical spaces, enabling a single model to be trained for all compounds and endpoints. We demonstrate improvements in accuracy on the latest chemistry and data when updating models with new data as an ongoing medicinal chemistry project progresses.

We present a general method called atom-wise free energy perturbation (AFEP), which extends a conventional molecular dynamics free energy perturbation (FEP) simulation to give the contribution to a free energy change from each atom. AFEP is derived from an expansion of the Zwanzig equation used in the exponential averaging method by defining that the system total energy can be partitioned into contributions from each atom. A partitioning method is assumed and used to group terms in the expansion to correspond to individual atoms. AFEP is applied to six example free energy changes to demonstrate the method. Firstly, the hydration free energies of methane, methanol, methylamine, methanethiol, and caffeine in water. AFEP highlights the atoms in the molecules that interact favorably or unfavorably with water. Finally AFEP is applied to the binding free energy of human immunodeficiency virus type 1 protease to lopinavir, and AFEP reveals the contribution of each atom to the binding free energy, indicating candidate areas of the molecule to improve to produce a more strongly binding inhibitor. FEP gives a single value for the free energy change and is already a very useful method. AFEP gives a free energy change for each "part" of the system being simulated, where part can mean individual atoms, chemical groups, amino acids, or larger partitions depending on what the user is trying to measure. This method should have various applications in molecular dynamics studies of physical, chemical, or biochemical phenomena, specifically in the field of computational drug discovery.