Generating Focused Molecule Libraries for Drug Discovery with Recurrent Neural Networks.
Ontology highlight
ABSTRACT:
SUBMITTER: Segler MHS
PROVIDER: S-EPMC5785775 | biostudies-literature | 2018 Jan
REPOSITORIES: biostudies-literature
Publications
Generating Focused Molecule Libraries for Drug Discovery with Recurrent Neural Networks.
Segler Marwin H S MHS Kogej Thierry T Tyrchan Christian C Waller Mark P MP
ACS central science 20171228 1
In <i>de novo</i> drug design, computational strategies are used to generate novel molecules with good affinity to the desired biological target. In this work, we show that recurrent neural networks can be trained as generative models for molecular structures, similar to statistical language models in natural language processing. We demonstrate that the properties of the generated molecules correlate very well with the properties of the molecules used to train the model. In order to enrich libra ...[more]
Similar Datasets
Project description:The discovery and development of novel pharmaceuticals is an area of active research mainly due to the large investments required and long payback times. As of 2016, the development of a novel drug candidate required up to $ USD 2.6 billion in investment for only 10% rate of approval by the FDA. To help decreasing the costs associated with the process, a number of in silico approaches have been developed with relatively low success due to limited predicting performance. Here, we introduced a machine learning-based algorithm as an alternative for a more accurate search of new pharmacological candidates, which takes advantage of Recurrent Neural Networks (RNN) for active molecule prediction within large databases. Our approach, termed PharmaNet was implemented here to search for ligands against specific cell receptors within 102 targets of the DUD-E database, which contains 22886 active molecules. PharmaNet comprises three main phases. First, a SMILES representation of the molecule is converted into a raw molecular image. Second, a convolutional encoder processes the data to obtain a fingerprint molecular image that is finally analyzed by a Recurrent Neural Network (RNN). This approach enables precise predictions of the molecules' target on the basis of the feature extraction, the sequence analysis and the relevant information filtered out throughout the process. Molecule Target prediction is a highly unbalanced detection problem and therefore, we propose that an adequate evaluation metric of performance is the area under the Normalized Average Precision (NAP) curve. PharmaNet largely surpasses the previous state-of-the-art method with 97.7% in the Receiver Operating Characteristic curve (ROC-AUC) and 65.5% in the NAP curve. We obtained a perfect performance for human farnesyl pyrophosphate synthase (FPPS), which is a potential target for antimicrobial and anticancer treatments. We decided to test PharmaNet for activity prediction against FPPS by searching in the CHEMBL data set. We obtained three (3) potential inhibitors that were further validated through both molecular docking and in silico toxicity prediction. Most importantly, one of this candidates, CHEMBL2007613, was predicted as a potential antiviral due to its involvement on the PCDH17 pathway, which has been reported to be related to viral infections.
| S-EPMC8075191 | biostudies-literature
Project description:A recurrent neural network (RNN) is a machine learning model that learns the relationship between elements of an input series, in addition to inferring a relationship between the data input to the model and target output. Memory augmentation allows the RNN to learn the interrelationships between elements of the input over a protracted length of the input series. Inspired by the success of stack augmented RNN (StackRNN) to generate strings for various applications, we present two memory augmented RNN-based architectures: the Neural Turing Machine (NTM) and the Differentiable Neural Computer (DNC) for the de-novo generation of small molecules. We trained a character-level convolutional neural network (CNN) to predict the properties of a generated string and compute a reward or loss in a deep reinforcement learning setup to bias the Generator to produce molecules with the desired property. Further, we compare the performance of these architectures to gain insight to their relative merits in terms of the validity and novelty of the generated molecules and the degree of property bias towards the computational generation of de-novo drugs. We also compare the performance of these architectures with simpler recurrent neural networks (Vanilla RNN, LSTM, and GRU) without an external memory component to explore the impact of augmented memory in the task of de-novo generation of small molecules.
| S-EPMC9223405 | biostudies-literature
Project description:The development of bioactive small molecules as probes or drug candidates requires discovery platforms that enable access to chemical diversity and can quickly reveal new ligands for a target of interest. Within the past 15 years, DNA-encoded library (DEL) technology has matured into a widely used platform for small-molecule discovery, yielding a wide variety of bioactive ligands for many therapeutically relevant targets. DELs offer many advantages compared with traditional screening methods, including efficiency of screening, easily multiplexed targets and library selections, minimized resources needed to evaluate an entire DEL and large library sizes. This Review provides accounts of recently described small molecules discovered from DELs, including their initial identification, optimization and validation of biological properties including suitability for clinical applications.
| S-EPMC10924799 | biostudies-literature
Project description:Research productivity in the pharmaceutical industry has declined significantly in recent decades, with higher costs, longer timelines, and lower success rates of drug candidates in clinical trials. This has prioritized the scalability and multiobjectivity of drug discovery and design. De novo drug design has emerged as a promising approach; molecules are generated from scratch, thus reducing the reliance on trial and error and premade molecular repositories. However, optimizing for molecular traits remains challenging, impeding the implementation of de novo methods. In this work, we propose a de novo approach capable of optimizing multiple traits collectively. A recurrent neural network was used to generate molecules which were then ranked based on multiple properties by a nondominated sorting algorithm. The best of the molecules generated were selected and used to fine-tune the recurrent neural network through transfer learning, creating a cycle that mimics the traditional design–synthesis–test cycle. We demonstrate the efficacy of this approach through a proof of concept, optimizing for constraints on molecular weight, octanol-water partition coefficient, the number of rotatable bonds, hydrogen bond donors, and hydrogen bond acceptors simultaneously. Analysis of the molecules generated after five iterations of the cycle revealed a 14-fold improvement in the quality of generated molecules, along with improvements to the accuracy of the recurrent neural network and the structural diversity of the molecules generated. This cycle notably does not require large amounts of training data nor any handwritten scoring functions. Altogether, this approach uniquely combines scalable generation with multiobjective optimization of molecules.
| S-EPMC7026957 | biostudies-literature
Project description:Analysis of compound-protein interactions (CPIs) has become a crucial prerequisite for drug discovery and drug repositioning. In vitro experiments are commonly used in identifying CPIs, but it is not feasible to discover the molecular and proteomic space only through experimental approaches. Machine learning's advances in predicting CPIs have made significant contributions to drug discovery. Deep neural networks (DNNs), which have recently been applied to predict CPIs, performed better than other shallow classifiers. However, such techniques commonly require a considerable volume of dense data for each training target. Although the number of publicly available CPI data has grown rapidly, public data is still sparse and has a large number of measurement errors. In this paper, we propose a novel method, Multi-channel PINN, to fully utilize sparse data in terms of representation learning. With representation learning, Multi-channel PINN can utilize three approaches of DNNs which are a classifier, a feature extractor, and an end-to-end learner. Multi-channel PINN can be fed with both low and high levels of representations and incorporates each of them by utilizing all approaches within a single model. To fully utilize sparse public data, we additionally explore the potential of transferring representations from training tasks to test tasks. As a proof of concept, Multi-channel PINN was evaluated on fifteen combinations of feature pairs to investigate how they affect the performance in terms of highest performance, initial performance, and convergence speed. The experimental results obtained indicate that the multi-channel models using protein features performed better than single-channel models or multi-channel models using compound features. Therefore, Multi-channel PINN can be advantageous when used with appropriate representations. Additionally, we pretrained models on a training task then finetuned them on a test task to figure out whether Multi-channel PINN can capture general representations for compounds and proteins. We found that there were significant differences in performance between pretrained models and non-pretrained models.
| S-EPMC6617572 | biostudies-literature
Project description:BackgroundTranscription factors regulate gene expression by binding to transcription factor binding sites (TFBSs). Most models for predicting TFBSs are based on position weight matrices (PWMs), which require a specific motif to be present in the DNA sequence and do not consider interdependencies of nucleotides. Novel approaches such as Transcription Factor Flexible Models or recurrent neural networks consequently provide higher accuracies. However, it is unclear whether such approaches can uncover novel non-canonical, hitherto unexpected TFBSs relevant to human transcriptional regulation.ResultsIn this study, we trained a convolutional recurrent neural network with HT-SELEX data for GRHL1 binding and applied it to a set of GRHL1 binding sites obtained from ChIP-Seq experiments from human cells. We identified 46 non-canonical GRHL1 binding sites, which were not found by a conventional PWM approach. Unexpectedly, some of the newly predicted binding sequences lacked the CNNG core motif, so far considered obligatory for GRHL1 binding. Using isothermal titration calorimetry, we experimentally confirmed binding between the GRHL1-DNA binding domain and predicted GRHL1 binding sites, including a non-canonical GRHL1 binding site. Mutagenesis of individual nucleotides revealed a correlation between predicted binding strength and experimentally validated binding affinity across representative sequences. This correlation was neither observed with a PWM-based nor another deep learning approach.ConclusionsOur results show that convolutional recurrent neural networks may uncover unanticipated binding sites and facilitate quantitative transcription factor binding predictions.
| S-EPMC10696883 | biostudies-literature
Project description:Neural circuits display complex activity patterns both spontaneously and when responding to a stimulus or generating a motor output. How are these two forms of activity related? We develop a procedure called FORCE learning for modifying synaptic strengths either external to or within a model neural network to change chaotic spontaneous activity into a wide variety of desired activity patterns. FORCE learning works even though the networks we train are spontaneously chaotic and we leave feedback loops intact and unclamped during learning. Using this approach, we construct networks that produce a wide variety of complex output patterns, input-output transformations that require memory, multiple outputs that can be switched by control inputs, and motor patterns matching human motion capture data. Our results reproduce data on premovement activity in motor and premotor cortex, and suggest that synaptic plasticity may be a more rapid and powerful modulator of network activity than generally appreciated.
| S-EPMC2756108 | biostudies-literature
Project description:The use of virtual drug screening can be beneficial to research teams, enabling them to narrow down potentially useful compounds for further study. A variety of virtual screening methods have been developed, typically with machine learning classifiers at the center of their design. In the present study, we created a virtual screener for protein kinase inhibitors. Experimental compound-target interaction data were obtained from the IDG-DREAM Drug-Kinase Binding Prediction Challenge. These data were converted and fed as inputs into two multi-input recurrent neural networks (RNNs). The first network utilized data encoded in one-hot representation, while the other incorporated embedding layers. The models were developed in Python, and were designed to output the IC50 of the target compounds. The performance of the models was assessed primarily through analysis of the Q2 values produced from runs of differing sample and epoch size; recorded loss values were also reported and graphed. The performance of the models was limited, though multiple changes are proposed for potential improvement of a multi-input recurrent neural network-based screening tool.
| S-EPMC7435591 | biostudies-literature
Project description:Chemical libraries have become of utmost importance to boost drug discovery processes. It is widely accepted that the quality of a chemical library depends, among others, on its availability and chemical diversity which help in rising the chances of finding good hits. In this regard, our group has developed a source for useful chemicals named Medicinal and Biological Chemistry (MBC) library. It originates from more than 30 years of experience in drug design and discovery of our research group and has successfully provided effective hits for neurological, neurodegenerative and infectious diseases. Moreover, in the last years, the European research infrastructure for chemical biology EU-OPENSCREEN has generated the European Chemical Biology library (ECBL) to be used as a source of hits for drug discovery. Here we present and discuss the updated version of the MBC library (MBC v.2022), enriched with new scaffolds and containing more than 2,500 compounds together with ECBL that collects about 100,000 small molecules. To properly address the improved potentialities of the new version of our MBC library in drug discovery, up to 44 among physicochemical and pharmaceutical properties have been calculated and compared with those of other well-known publicly available libraries. For comparison, we have used ZINC20, DrugBank, ChEMBL library, ECBL and NuBBE along with an approved drug library. Final results allowed to confirm the competitive chemical space covered by MBC v.2022 and ECBL together with suitable drug-like properties. In all, we can affirm that these two libraries represent an interesting source of new hits for drug discovery.
| S-EPMC10457160 | biostudies-literature
Project description:Syndromic surveillance detects and monitors individual and population health indicators through sources such as emergency department records. Automated classification of these records can improve outbreak detection speed and diagnosis accuracy. Current syndromic systems rely on hand-coded keyword-based methods to parse written fields and may benefit from the use of modern supervised-learning classifier models. In this paper, we implement two recurrent neural network models based on long short-term memory (LSTM) and gated recurrent unit (GRU) cells and compare them to two traditional bag-of-words classifiers: multinomial naïve Bayes (MNB) and a support vector machine (SVM). The MNB classifier is one of only two machine learning algorithms currently being used for syndromic surveillance. All four models are trained to predict diagnostic code groups as defined by Clinical Classification Software, first to predict from discharge diagnosis, and then from chief complaint fields. The classifiers are trained on 3.6 million de-identified emergency department records from a single United States jurisdiction. We compare performance of these models primarily using the F1 score, and we measure absolute model performance to determine which conditions are the most amenable to surveillance based on chief complaint alone. Using discharge diagnoses, the LSTM classifier performs best, though all models exhibit an F1 score above 96.00. Using chief complaints, the GRU performs best (F1 = 47.38), and MNB with bigrams performs worst (F1 = 39.40). We also note that certain syndrome types are easier to detect than others. For example, chief complaints using the GRU model predicts alcohol-related disorders well (F1 = 78.91) but predicts influenza poorly (F1 = 14.80). In all instances, the RNN models outperformed the bag-of-words classifiers suggesting deep learning models could substantially improve the automatic classification of unstructured text for syndromic surveillance.
| S-EPMC10563436 | biostudies-literature