Search

The vast size of chemical space necessitates computational approaches to automate and accelerate the design of molecular sequences to guide experimental efforts for drug discovery. Genetic algorithms provide a useful framework to incrementally generate molecules by applying mutations to known chemical structures. Recently, masked language models have been applied to automate the mutation process by leveraging large compound libraries to learn commonly occurring chemical sequences (i.e., using tokenization) and predict rearrangements (i.e., using mask prediction). Here, we consider how language models can be adapted to improve molecule generation for different optimization tasks. We use two different generation strategies for comparison, fixed and adaptive. The fixed strategy uses a pre-trained model to generate mutations; the adaptive strategy trains the language model on each new generation of molecules selected for target properties during optimization. Our results show that the adaptive strategy allows the language model to more closely fit the distribution of molecules in the population. Therefore, for enhanced fitness optimization, we suggest the use of the fixed strategy during an initial phase followed by the use of the adaptive strategy. We demonstrate the impact of adaptive training by searching for molecules that optimize both heuristic metrics, drug-likeness and synthesizability, as well as predicted protein binding affinity from a surrogate model. Our results show that the adaptive strategy provides a significant improvement in fitness optimization compared to the fixed pre-trained model, empowering the application of language models to molecular design tasks

Efficient methods for searching the chemical space of molecular compounds are needed to automate and accelerate the design of new functional molecules such as pharmaceuticals. Given the high cost in both resources and time for experimental efforts, computational approaches play a key role in guiding the selection of promising molecules for further investigation. Here, we construct a workflow to accelerate design by combining approximate quantum chemical methods [i.e. density-functional tight-binding (DFTB)], a graph convolutional neural network (GCNN) surrogate model for chemical property prediction, and a masked language model (MLM) for molecule generation. Property data from the DFTB calculations are used to train the surrogate model; the surrogate model is used to score candidates generated by the MLM. The surrogate reduces computation time by orders of magnitude compared to the DFTB calculations, enabling an increased search of chemical space. Furthermore, the MLM generates a diverse set of chemical modifications based on pre-training from a large compound library. We utilize the workflow to search for near-infrared photoactive molecules by minimizing the predicted HOMO-LUMO gap as the target property. Our results show that the workflow can generate optimized molecules outside of the original training set, which suggests that iterations of the workflow could be useful for searching vast chemical spaces in a wide range of design problems.

We aim to reduce overfitting and model overconfidence by distilling the knowledge of an ensemble of deep learning models into a single model for the classification of cancer pathology reports.We consider the text classification problem that involves 5 individual tasks. The baseline model consists of a multitask convolutional neural network (MtCNN), and the implemented ensemble (teacher) consists of 1000 MtCNNs. We performed knowledge transfer by training a single model (student) with soft labels derived through the aggregation of ensemble predictions. We evaluate performance based on accuracy and abstention rates by using softmax thresholding.The student model outperforms the baseline MtCNN in terms of abstention rates and accuracy, thereby allowing the model to be used with a larger volume of documents when deployed. The highest boost was observed for subsite and histology, for which the student model classified an additional 1.81% reports for subsite and 3.33% reports for histology.Ensemble predictions provide a useful strategy for quantifying the uncertainty inherent in labeled data and thereby enable the construction of soft labels with estimated probabilities for multiple classes for a given document. Training models with the derived soft labels reduce model confidence in difficult-to-classify documents, thereby leading to a reduction in the number of highly confident wrong predictions.Ensemble model distillation is a simple tool to reduce model overconfidence in problems with extreme class imbalance and noisy datasets. These methods can facilitate the deployment of deep learning models in high-risk domains with low computational resources where minimizing inference time is required.

The objective of this work is to develop a biorelevant dissolution method to support the clinical study for In Vitro In Vivo Correlation (IVIVC) of the first commercially approved single-layer extrudable core system (ECS) osmotic tablet - the 11 mg tofacitinib modified-release tablet. The dissolution conditions were selected through analysis of experimental work including several designed experiments (DoE). The Apparatus 2 (paddles) was selected over the Apparatus 1 (baskets) to minimize the dissolution test variability. The paddle speed was kept at 50 rpm to be conservative and because higher paddle speed did not offer statistically significant improvement in dissolution test variability. The buffer of 50 mM potassium phosphate at pH 6.8 was selected over other buffers at lower or acid pH as the in vivo drug release is expected to occur in the small intestinal region, where the pH is approximately neutral. Finally, the statistically designed experiments proved that use of the Japanese basket sinkers was effective in reducing dissolution variability and eliminating the artificial shift in dissolution profile caused by final pink color-coated tablets sticking to the dissolution vessel. Discriminatory power of the method was verified and the method was validated per ICH and FDA guidelines. Since a Level A IVIVC is established from the analysis of the results of both in vivo clinical study and in vitro dissolution testing, the method is proven to be biorelevant. It also serves a suitable quality control dissolution method.

Purpose To determine if a validated Level A in-vitro in-vivo correlation (IVIVC) could be achieved with the extrudable core system (ECS) osmotic tablet platform. Tofacitinib is an oral JAK inhibitor for the treatment of rheumatoid arthritis. Methods Fast-, medium-, and slow-release modified-release formulations of 11 mg tofacitinib ECS tablets, and one formulation of 22 mg tofacitinib ECS tablet, were manufactured. In vitro dissolution of the tofacitinib ECS tablets was performed using USP Apparatus 2 (paddles) and in vivo pharmacokinetic (PK) data were obtained from a Phase 1 study in healthy volunteers. A 5 mg immediate-release formulation tablet was included to support deconvolution of the tofacitinib ECS PK tablet data to obtain the in vivo absorption profiles. A linear, piecewise correlation and a simple linear correlation were used to build and validate two IVIVC models. Results The prediction errors (PEs) for the linear, piecewise correlation met the Food and Drug Administration's criteria for establishing a Level A IVIVC, with a maximum absolute individual internal PE of 4.6%, a maximum absolute average internal PE of 3.9%, and a maximum absolute external PE of 8.4% obtained. Conclusions This study demonstrates that the tofacitinib ECS osmotic tablet platform can achieve a Level A IVIVC, similar to other osmotic delivery systems.

In this paper, the authors present a wavelet-based algorithm (Wave-SOM) to help visualize and cluster oscillatory time-series data in two-dimensional gene expression micro-arrays. Using various wavelet transformations, raw data are first de-noised by decomposing the time-series into low and high frequency wavelet coefficients. Following thresholding, the coefficients are fed as an input vector into a two-dimensional Self-Organizing-Map clustering algorithm. Transformed data are then clustered by minimizing the Euclidean (L2) distance between their corresponding fluctuation patterns. A multi-resolution analysis by Wave-SOM of expression data from the yeast Saccharomyces cerevisiae, exposed to oxidative stress and glucose-limited growth, identified 29 genes with correlated expression patterns that were mapped into 5 different nodes. The ordered clustering of yeast genes by Wave-SOM illustrates that the same set of genes (encoding ribosomal proteins) can be regulated by two different environmental stresses, oxidative stress and starvation. The algorithm provides heuristic information regarding the similarity of different genes. Using previously studied expression patterns of yeast cell-cycle and functional genes as test data sets, the authors’ algorithm outperformed five other competing programs.

A multifaceted approach was successfully used to identify three of four unknown degradants in degraded low dose tablets. Accelerated solvent extraction (ASE) was found to be an invaluable tool in this multifaceted approach. ASE was capable of extracting four individual degradants of an active pharmaceutical component from 10 tablets into 15 mL of solvent with approximately 100% recovery for each degradant. Using ASE instead of manual extraction led to the extraction and isolation of the degradants in 1 day instead of 7 days. One of the degradants was extracted by ASE, isolated by semi-prep HPLC, and identified by LC-MS and NMR spectroscopy. The structures of two of the remaining three degradants were confirmed by synthesis of authentic samples, while the fourth degradant is yet to be identified.