RetroGNN: Fast Estimation of Synthesizability for Virtual Screening and De Novo Design by Learning from Slow Retrosynthesis Software.

De novo molecule design algorithms often result in chemically unfeasible or synthetically inaccessible molecules. A natural idea to mitigate this problem is to bias these algorithms toward more easily synthesizable molecules using a proxy score for synthetic accessibility. However, using currently available proxies can still result in highly unrealistic compounds. Here, we propose a novel approach, RetroGNN, to estimate synthesizability. First, we search for routes using synthesis planning software for a large number of random molecules. This information is then used to train a graph neural network to predict the outcome of the synthesis planner given the target molecule, in which the regression task can be used as a synthesizability scorer. We highlight how RetroGNN can be used in generative molecule-discovery pipelines together with other scoring functions. We evaluate our approach on several QSAR-based molecule design benchmarks, for which we find synthesizable molecules with state-of-the-art scores. Compared to the virtual screening of 5 million existing molecules from the ZINC database, using RetroGNNScore with a simple fragment-based de novo design algorithm finds molecules predicted to be more likely to possess the desired activity exponentially faster, while maintaining good druglike properties and being easier to synthesize. Importantly, our deep neural network can successfully filter out hard to synthesize molecules while achieving a 105 times speedup over using retrosynthesis planning software.

Molecule Edit Graph Attention Network: Modeling Chemical Reactions as Sequences of Graph Edits.

The central challenge in automated synthesis planning is to be able to generate and predict outcomes of a diverse set of chemical reactions. In particular, in many cases, the most likely synthesis pathway cannot be applied due to additional constraints, which requires proposing alternative chemical reactions. With this in mind, we present Molecule Edit Graph Attention Network (MEGAN), an end-to-end encoder-decoder neural model. MEGAN is inspired by models that express a chemical reaction as a sequence of graph edits, akin to the arrow pushing formalism. We extend this model to retrosynthesis prediction (predicting substrates given the product of a chemical reaction) and scale it up to large data sets. We argue that representing the reaction as a sequence of edits enables MEGAN to efficiently explore the space of plausible chemical reactions, maintaining the flexibility of modeling the reaction in an end-to-end fashion and achieving state-of-the-art accuracy in standard benchmarks. Code and trained models are made available online at https://github.com/molecule-one/megan.

Emulating Docking Results Using a Deep Neural Network: A New Perspective for Virtual Screening.

Docking is one of the most important steps in virtual screening pipelines, and it is an established method for examining potential interactions between ligands and receptors. However, this method is computationally expensive, and it is often among the last steps of the process of compound libraries evaluation. In this work, we investigate the feasibility of learning a deep neural network to predict the docking output directly from a two-dimensional compound structure. The developed protocol is orders of magnitude faster than typical docking software, and it returns ligand-receptor complexes encoded in the form of the interaction fingerprint. Its speed and efficiency unlock the application possibilities, such as screening compound libraries of vast size on the basis of contact patterns or docking score (derived on the basis of predicted interaction schemes). We tested our approach on several G protein-coupled receptor targets and 4 CYP enzymes in retrospective virtual screening experiments, and a variant of graph convolutional network appeared to be most effective in emulating docking results. The method can be easily used by the community based on the code available in the Supporting Information.

Development of New Methods Needs Proper Evaluation-Benchmarking Sets for Machine Learning Experiments for Class A GPCRs.

New computational approaches for virtual screening applications are constantly being developed. However, before a particular tool is used to search for new active compounds, its effectiveness in the type of task must be examined. In this study, we conducted a detailed analysis of various aspects of preparation of respective data sets for such an evaluation. We propose a protocol for fetching data from the ChEMBL database, examine various compound representations in terms of the possible bias resulting from the way they are generated, and define a new metric for comparing the structural similarity of compounds, which is in line with chemical intuition. The newly developed method is also used for the evaluation of various approaches for division of the data set into training and test set parts, which are also examined in detail in terms of being the source of possible results bias. Finally, machine learning methods are applied in cross-validation studies of data sets constructed within the paper, constituting benchmarks for the assessment of computational methods developed for virtual screening tasks. Additionally, analogous data sets for class A G protein-coupled receptors (100 targets with the highest number of records) were prepared. They are available at http://gmum.net/benchmarks/ , together with script enabling reproduction of all results available at https://github.com/lesniak43/ananas .

Three-dimensional descriptors for aminergic GPCRs: dependence on docking conformation and crystal structure.

Three-dimensional descriptors are often used to search for new biologically active compounds, in both ligand- and structure-based approaches, capturing the spatial orientation of molecules. They frequently constitute an input for machine learning-based predictions of compound activity or quantitative structure-activity relationship modeling; however, the distribution of their values and the accuracy of depicting compound orientations might have an impact on the power of the obtained predictive models. In this study, we analyzed the distribution of three-dimensional descriptors calculated for docking poses of active and inactive compounds for all aminergic G protein-coupled receptors with available crystal structures, focusing on the variation in conformations for different receptors and crystals. We demonstrated that the consistency in compound orientation in the binding site is rather not correlated with the affinity itself, but is more influenced by other factors, such as the number of rotatable bonds and crystal structure used for docking studies. The visualizations of the descriptors distributions were prepared and made available online at http://chem.gmum.net/vischem_stability , which enables the investigation of chemical structures referring to particular data points depicted in the figures. Moreover, the performed analysis can assist in choosing crystal structure for docking studies, helping in selection of conditions providing the best discrimination between active and inactive compounds in machine learning-based experiments.

Quo vadis G protein-coupled receptor ligands? A tool for analysis of the emergence of new groups of compounds over time.

Exponential growth in the number of compounds with experimentally verified activity towards particular target has led to the emergence of various databases gathering data on biological activity. In this study, the ligands of family A of the G Protein-Coupled Receptors that are collected in the ChEMBL database were examined, and special attention was given to serotonin receptors. Sets of compounds were examined in terms of their appearance over time, they were mapped to the chemical space of drugs deposited in DrugBank, and the emergence of structurally new clusters of compounds was indicated. In addition, a tool for detailed analysis of the obtained visualizations was prepared and made available online at http://chem.gmum.net/vischem, which enables the investigation of chemical structures while referring to particular data points depicted in the figures and changes in compounds datasets over time.

Relative molecule self-attention transformer.

The prediction of molecular properties is a crucial aspect in drug discovery that can save a lot of money and time during the drug design process. The use of machine learning methods to predict molecular properties has become increasingly popular in recent years. Despite advancements in the field, several challenges remain that need to be addressed, like finding an optimal pre-training procedure to improve performance on small datasets, which are common in drug discovery. In our paper, we tackle these problems by introducing Relative Molecule Self-Attention Transformer for molecular representation learning. It is a novel architecture that uses relative self-attention and 3D molecular representation to capture the interactions between atoms and bonds that enrich the backbone model with domain-specific inductive biases. Furthermore, our two-step pretraining procedure allows us to tune only a few hyperparameter values to achieve good performance comparable with state-of-the-art models on a wide selection of downstream tasks.

An Efficient Deep Neural Network to Classify Large 3D Images With Small Objects.

3D imaging enables accurate diagnosis by providing spatial information about organ anatomy. However, using 3D images to train AI models is computationally challenging because they consist of 10x or 100x more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to 2D. We propose an effective alternative, a neural network that enables efficient classification of full-resolution 3D medical images. Compared to off-the-shelf convolutional neural networks, our network, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% less computation. While it is trained only with image-level labels, without segmentation labels, it explains its predictions by providing pixel-level saliency maps. On a dataset collected at NYU Langone Health, including 85,526 patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and 3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) in classifying breasts with malignant findings using 3D mammography. This is comparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) and synthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMIC successfully classified large 3D images despite focusing computation on a smaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifies and utilizes extremely small regions of interest from 3D images consisting of hundreds of millions of pixels, dramatically reducing associated computational challenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from Duke University Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

Differences between human and machine perception in medical diagnosis.

Deep neural networks (DNNs) show promise in image-based medical diagnosis, but cannot be fully trusted since they can fail for reasons unrelated to underlying pathology. Humans are less likely to make such superficial mistakes, since they use features that are grounded on medical science. It is therefore important to know whether DNNs use different features than humans. Towards this end, we propose a framework for comparing human and machine perception in medical diagnosis. We frame the comparison in terms of perturbation robustness, and mitigate Simpson's paradox by performing a subgroup analysis. The framework is demonstrated with a case study in breast cancer screening, where we separately analyze microcalcifications and soft tissue lesions. While it is inconclusive whether humans and DNNs use different features to detect microcalcifications, we find that for soft tissue lesions, DNNs rely on high frequency components ignored by radiologists. Moreover, these features are located outside of the region of the images found most suspicious by radiologists. This difference between humans and machines was only visible through subgroup analysis, which highlights the importance of incorporating medical domain knowledge into the comparison.

An artificial intelligence system for predicting the deterioration of COVID-19 patients in the emergency department.

During the coronavirus disease 2019 (COVID-19) pandemic, rapid and accurate triage of patients at the emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction of deterioration risk using a deep neural network that learns from chest X-ray images and a gradient boosting model that learns from routine clinical variables. Our AI prognosis system, trained using data from 3661 patients, achieves an area under the receiver operating characteristic curve (AUC) of 0.786 (95% CI: 0.745-0.830) when predicting deterioration within 96 hours. The deep neural network extracts informative areas of chest X-ray images to assist clinicians in interpreting the predictions and performs comparably to two radiologists in a reader study. In order to verify performance in a real clinical setting, we silently deployed a preliminary version of the deep neural network at New York University Langone Health during the first wave of the pandemic, which produced accurate predictions in real-time. In summary, our findings demonstrate the potential of the proposed system for assisting front-line physicians in the triage of COVID-19 patients.

Improving the Ability of Deep Neural Networks to Use Information from Multiple Views in Breast Cancer Screening.

In breast cancer screening, radiologists make the diagnosis based on images that are taken from two angles. Inspired by this, we seek to improve the performance of deep neural networks applied to this task by encouraging the model to use information from both views of the breast. First, we took a closer look at the training process and observed an imbalance between learning from the two views. In particular, we observed that layers processing one of the views have parameters with larger gradients in magnitude, and contribute more to the overall loss reduction. Next, we tested several methods targeted at utilizing both views more equally in training. We found that using the same weights to process both views, or using modality dropout, leads to a boost in performance. Looking forward, our results indicate improving learning dynamics as a promising avenue for improving utilization of multiple views in deep neural networks for medical diagnosis.

An artificial intelligence system for predicting the deterioration of COVID-19 patients in the emergency department.

During the coronavirus disease 2019 (COVID-19) pandemic, rapid and accurate triage of patients at the emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction of deterioration risk using a deep neural network that learns from chest X-ray images and a gradient boosting model that learns from routine clinical variables. Our AI prognosis system, trained using data from 3,661 patients, achieves an area under the receiver operating characteristic curve (AUC) of 0.786 (95% CI: 0.745-0.830) when predicting deterioration within 96 hours. The deep neural network extracts informative areas of chest X-ray images to assist clinicians in interpreting the predictions and performs comparably to two radiologists in a reader study. In order to verify performance in a real clinical setting, we silently deployed a preliminary version of the deep neural network at New York University Langone Health during the first wave of the pandemic, which produced accurate predictions in real-time. In summary, our findings demonstrate the potential of the proposed system for assisting front-line physicians in the triage of COVID-19 patients.

Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening.

We present a deep convolutional neural network for breast cancer screening exam classification, trained, and evaluated on over 200000 exams (over 1000000 images). Our network achieves an AUC of 0.895 in predicting the presence of cancer in the breast, when tested on the screening population. We attribute the high accuracy to a few technical advances. 1) Our network's novel two-stage architecture and training procedure, which allows us to use a high-capacity patch-level network to learn from pixel-level labels alongside a network learning from macroscopic breast-level labels. 2) A custom ResNet-based network used as a building block of our model, whose balance of depth and width is optimized for high-resolution medical images. 3) Pretraining the network on screening BI-RADS classification, a related task with more noisy labels. 4) Combining multiple input views in an optimal way among a number of possible choices. To validate our model, we conducted a reader study with 14 readers, each reading 720 screening mammogram exams, and show that our model is as accurate as experienced radiologists when presented with the same data. We also show that a hybrid model, averaging the probability of malignancy predicted by a radiologist with a prediction of our neural network, is more accurate than either of the two separately. To further understand our results, we conduct a thorough analysis of our network's performance on different subpopulations of the screening population, the model's design, training procedure, errors, and properties of its internal representations. Our best models are publicly available at https://github.com/nyukat/breast_cancer_classifier.