Recurrent Neural Networks (RNNs) trained with a set of molecules represented as unique (canonical) SMILES strings, have shown the capacity to create large chemical spaces valid and meaningful structures. Herein we perform an extensive benchmark on models subsets GDB-13 different sizes (1 million, 10,000 1000), variants (canonical, randomized DeepSMILES), two recurrent cell types (LSTM GRU) hyperparameter combinations. To guide benchmarks new metrics were developed that define how well model...

Deep learning methods applied to drug discovery have been used generate novel structures. In this study, we propose a new deep architecture, LatentGAN, which combines an autoencoder and generative adversarial neural network for de novo molecular design. We the method in two scenarios: one random drug-like compounds another target-biased compounds. Our results show that works well both cases. Sampled from trained model can largely occupy same chemical space as training set also substantial...

Simplified Molecular Input Line Entry System (SMILES) is a single line text representation of unique molecule. One molecule can however have multiple SMILES strings, which reason that canonical been defined, ensures one to correspondence between string and Here the fact represent same explored as technique for data augmentation molecular QSAR dataset modeled by long short term memory (LSTM) cell based neural network. The augmented was 130 times bigger than original. network trained with...

We present the open-source AiZynthFinder software that can be readily used in retrosynthetic planning. The algorithm is based on a Monte Carlo tree search recursively breaks down molecule to purchasable precursors. guided by an artificial neural network policy suggests possible precursors utilizing library of known reaction templates. fast and typically find solution less than 10 s perform complete 1 min. Moreover, development code was range engineering principles such as automatic testing,...

Chemical autoencoders are attractive models as they combine chemical space navigation with possibilities for de novo molecule generation in areas of interest. This enables them to produce focused libraries around a single lead compound employment early drug discovery project. Here, it is shown that the choice representation, such strings from simplified molecular-input line-entry system (SMILES), has large influence on properties latent space. It further explored what extent translating...

Molecular generative models trained with small sets of molecules represented as SMILES strings can generate large regions the chemical space. Unfortunately, due to sequential nature strings, these are not able given a scaffold (i.e., partially-built explicit attachment points). Herein we report new SMILES-based molecular architecture that generates from scaffolds and be any arbitrary set. This approach is possible thanks set pre-processing algorithm exhaustively slices all combinations...

Abstract Transformer models coupled with a simplified molecular line entry system (SMILES) have recently proven to be powerful combination for solving challenges in cheminformatics. These models, however, are often developed specifically single application and can very resource-intensive train. In this work we present the Chemformer model—a Transformer-based model which quickly applied both sequence-to-sequence discriminative cheminformatics tasks. Additionally, show that self-supervised...

The retrosynthetic accessibility score (RAscore) is based on AI driven planning, and useful for rapid scoring of synthetic feasability pre-screening large datasets virtual/generated molecules.

Here we show that Generative Topographic Mapping (GTM) can be used to explore the latent space of SMILES-based autoencoders and generate focused molecular libraries interest. We have built a sequence-to-sequence neural network with Bidirectional Long Short-Term Memory layers trained it on SMILES strings from ChEMBL23. Very high reconstruction rates test set molecules were achieved (>98%), which are comparable ones reported in related publications. Using GTM, visualized autoencoder...

Computer Assisted Synthesis Planning (CASP), datasets and their performance.

Abstract Deep learning methods applied to chemistry can be used accelerate the discovery of new molecules. This work introduces GraphINVENT, a platform developed for graph-based molecular design using graph neural networks (GNNs). GraphINVENT uses tiered deep network architecture probabilistically generate molecules single bond at time. All models implemented in quickly learn build resembling training set without any explicit programming chemical rules. The have been benchmarked MOSES...

Deep learning methods are used on spectroscopic data to predict drug content in tablets from near infrared (NIR) spectra. Using convolutional neural networks (CNNs), features ex- tracted the data. Extended multiplicative scatter correction (EMSC) and a novel spectral augmentation method benchmarked as preprocessing steps. The learned models perform better or par with hypothetical optimal partial least squares (PLS) for all combinations of preprocessing. Data subsequent EMSC combination gave...

Abstract A main challenge in drug discovery is finding molecules with a desirable balance of multiple properties. Here, we focus on the task molecular optimization, where goal to optimize given starting molecule towards This can be framed as machine translation problem natural language processing, our case, translated into optimized properties based SMILES representation. Typically, chemists would use their intuition suggest chemical transformations for being optimized. widely used strategy...

Molecular optimization aims to improve the drug profile of a starting molecule. It is fundamental problem in discovery but challenging due (i) requirement simultaneous multiple properties and (ii) large chemical space explore. Recently, deep learning methods have been proposed solve this task by mimicking chemist's intuition terms matched molecular pairs (MMPs). Although MMPs widely used strategy medicinal chemists, it offers limited capability exploring structural modifications, therefore...

The potential number of drug like small molecules is estimated to be between 10^23 and 10^60 while current databases known compounds are orders magnitude smaller with approximately 10^8 compounds. This discrepancy has led an interest in generating virtual libraries using hand crafted chemical rules fragment based methods cover a larger area space generate for use silico discovery endeavors. Here it explored what extent recurrent neural network long short term memory cells can figure out...

We present the open-source AiZynthFinder software that can be readily used in retrosynthetic planning. The algorithm is based on a Monte Carlo tree search recursively breaks down molecule to purchasable precursors. guided by an artificial neural network policy suggests possible precursors utilizing library of known reaction templates. fast and typically find solution less than 10 seconds perform complete 1 minute. Moreover, writing code was range engineering principles such as automatic...

PaRoutes is a framework benchmarking multi-step retrosynthesis methods. It consists of synthetic routes extracted from the patent literature, stock compounds, as well scripts to compute route quality and diversity metrics.

Recurrent Neural Networks (RNNs) trained with a set of molecules represented as unique (canonical) SMILES strings, have shown the capacity to create large chemical spaces valid and meaningful structures. Herein we perform an extensive benchmark on models subsets GDB-13 different sizes (1 million , 10,000 1,000), variants (canonical, randomized DeepSMILES), two recurrent cell types (LSTM GRU) hyperparameter combinations. To guide benchmarks new metrics were developed that define generated...

<p> </p><p>Deep learning methods applied to drug discovery have been used generate novel structures. In this study, we propose a new deep architecture, LatentGAN, which combines an autoencoder and generative adversarial neural network for de novo molecular design. We the method in two scenarios: one random drug-like compounds another target-biased compounds. Our results show that works well both cases: sampled from trained model can largely occupy same chemical space as...

Transformer models coupled with Simplified Molecular Line Entry System (SMILES) have recently proven to be a powerful combination for solving challenges in cheminformatics. These models, however, are often developed specifically single application and can very resource-intensive train. In this work we present Chemformer model – Transformerbased which quickly applied both sequence-to-sequence discriminative cheminformatics tasks. Additionally, show that self-supervised pre-training improve...

A de novo molecular design workflow can be used together with technologies such as reinforcement learning to navigate the chemical space. bottleneck in that remains solved is how integrate human feedback exploration of space optimize molecules. drug designer still needs goal, expressed a scoring function for molecules captures designer's implicit knowledge about optimization task. Little support this task exists and, consequently, chemist usually resorts iteratively building objective...