Integrating structure-based approaches in generative molecular design

Generative molecular design for drug discovery and development has seen a recent resurgence promising to improve the efficiency of the design-make-test-analyse cycle; by computationally exploring much larger chemical spaces than traditional virtual screening techniques. However, most generative models thus far have only utilized small-molecule information to train and condition de novo molecule generators. Here, we instead focus on recent approaches that incorporate protein structure into de novo molecule optimization in an attempt to maximize the predicted on-target binding affinity of generated molecules. We summarize these structure integration principles into either distribution learning or goal-directed optimization and for each case whether the approach is protein structure-explicit or implicit with respect to the generative model. We discuss recent approaches in the context of this categorization and provide our perspective on the future direction of the field.     

Artificial intelligence in molecular de novo design: Integration with experiment

In this mini review, we capture the latest progress of applying artificial intelligence (AI) techniques based on deep learning architectures to molecular de novo design with a focus on integration with experimental validation. We will cover the progress and experimental validation of novel generative algorithms, the validation of QSAR models and how AI-based molecular de novo design is starting to become connected with chemistry automation. While progress has been made in the last few years, it is still early days. The experimental validations conducted thus far should be considered proof-of-principle, providing confidence that the field is moving in the right direction.     

Isomeric lipid signatures reveal compartmentalized fatty acid metabolism in cancer

The cellular energy and biomass demands of cancer drive a complex dynamic between uptake of extracellular FAs and their de novo synthesis. Given that oxidation of de novo synthesized FAs for energy would result in net-energy loss, there is an implication that FAs from these two sources must have distinct metabolic fates; however, hitherto, all FAs have been considered part of a common pool. To probe potential metabolic partitioning of cellular FAs, cancer cells were supplemented with stable isotope-labeled FAs. Structural analysis of the resulting glycerophospholipids revealed that labeled FAs from uptake were largely incorporated to canonical (sn-) positions on the glycerol backbone. Surprisingly, labeled FA uptake also disrupted canonical isomer patterns of the unlabeled lipidome and induced repartitioning of n-3 and n-6 PUFAs into glycerophospholipid classes. These structural changes support the existence of differences in the metabolic fates of FAs derived from uptake or de novo sources and demonstrate unique signaling and remodeling behaviors usually hidden from conventional lipidomics.     

Deep generative models for 3D molecular structure

Deep generative models have gained recent popularity for chemical design. Many of these models have historically operated in 2D space; however, more recently explicit 3D molecular generative models have become of interest, which are the topic of this article. Dozens of published models have been developed in the last few years to generate molecules directly in 3D, outputting both the atom types and coordinates, either in one-shot or adding atoms or fragments step-by-step. These 3D generative models can also be guided by structural information such as a binding pocket representation to successfully generate molecules with docking score ranges similar to known actives, but still showing lower computational efficiency and generation throughput than 1D/2D generative models and sometimes producing unrealistic conformations. We advocate for a unified benchmark of metrics to evaluate generation and propose perspectives to be addressed in next implementations.     

Two-hit wonders: The expanding universe of multitargeting epigenetic agents

Multitargeting involves the application of molecules that are deliberately intended to bind to two or more unrelated cellular targets with high affinity. In epigenetic chemical biology and drug discovery, the rational design of multitargeting agents has evolved to a sophisticated level, and there are now five examples that have reached clinical trials. This review covers recent developments in the field and future prospects.     

Uncovering post-translational modification-associated protein–protein interactions

In living systems, the chemical space and functional repertoire of proteins are dramatically expanded through the post-translational modification (PTM) of various amino acid residues. These modifications frequently trigger unique protein–protein interactions (PPIs) – for example with reader proteins that directly bind the modified amino acid residue – which leads to downstream functional outcomes. The modification of a protein can also perturb its PPI network indirectly, for example, through altering its conformation or subcellular localization. Uncovering the network of unique PTM-triggered PPIs is essential to fully understand the roles of an ever-expanding list of PTMs in our biology. In this review, we discuss established strategies and current challenges associated with this endeavor.     

Meniscal lesion detection and characterization in adult knee MRI: A deep learning model approach with external validation

PurposeEvaluation of a deep learning approach for the detection of meniscal tears and their characterization (presence/absence of migrated meniscal fragment).MethodsA large annotated adult knee MRI database was built combining medical expertise of radiologists and data scientists’ tools. Coronal and sagittal proton density fat suppressed-weighted images of 11,353 knee MRI examinations (10,401 individual patients) paired with their standardized structured reports were retrospectively collected. After database curation, deep learning models were trained and validated on a subset of 8058 examinations. Algorithm performance was evaluated on a test set of 299 examinations reviewed by 5 musculoskeletal specialists and compared to general radiologists’ reports. External validation was performed using the publicly available MRNet database. Receiver Operating Characteristic (ROC) curves results and Area Under the Curve (AUC) values were obtained on internal and external databases.ResultsA combined architecture of meniscal localization and lesion classification 3D convolutional neural networks reached AUC values of 0.93 (95% CI 0.82, 0.95) for medial and 0.84 (95% CI 0.78, 0.89) for lateral meniscal tear detection, and 0.91 (95% CI 0.87, 0.94) for medial and 0.95 (95% CI 0.92, 0.97) for lateral meniscal tear migration detection. External validation of the combined medial and lateral meniscal tear detection models resulted in an AUC of 0.83 (95% CI 0.75, 0.90) without further training and 0.89 (95% CI 0.82, 0.95) with fine tuning.ConclusionOur deep learning algorithm demonstrated high performance in knee menisci lesion detection and characterization, validated on an external database.     

Deep generative modeling for volume reconstruction in cryo-electron microscopy

Advances in cryo-electron microscopy (cryo-EM) for high-resolution imaging of biomolecules in solution have provided new challenges and opportunities for algorithm development for 3D reconstruction. Next-generation volume reconstruction algorithms that combine generative modelling with end-to-end unsupervised deep learning techniques have shown promise, but many technical and theoretical hurdles remain, especially when applied to experimental cryo-EM images. In light of the proliferation of such methods, we propose here a critical review of recent advances in the field of deep generative modelling for cryo-EM reconstruction. The present review aims to (i) provide a unified statistical framework using terminology familiar to machine learning researchers with no specific background in cryo-EM, (ii) review the current methods in this framework, and (iii) outline outstanding bottlenecks and avenues for improvements in the field.     

The origin of long-chain fatty acids required for de novo ether lipid/plasmalogen synthesis

Peroxisomes are single-membrane bounded organelles that in humans play a dual role in lipid metabolism, including the degradation of very long-chain fatty acids and the synthesis of ether lipids/plasmalogens. The first step in de novo ether lipid synthesis is mediated by the peroxisomal enzyme glyceronephosphate O-acyltransferase, which has a strict substrate specificity reacting only with the long-chain acyl-CoAs. The aim of this study was to determine the origin of these long-chain acyl-CoAs. To this end, we developed a sensitive method for the measurement of de novo ether phospholipid synthesis in cells and, by CRISPR-Cas9 genome editing, generated a series of HeLa cell lines with deficiencies of proteins involved in peroxisomal biogenesis, beta-oxidation, ether lipid synthesis, or metabolite transport. Our results show that the long-chain acyl-CoAs required for the first step of ether lipid synthesis can be imported from the cytosol by the peroxisomal ABCD proteins, in particular ABCD3. Furthermore, we show that these acyl-CoAs can be produced intraperoxisomally by chain shortening of CoA esters of very long-chain fatty acids via beta-oxidation. Our results demonstrate that peroxisomal beta-oxidation and ether lipid synthesis are intimately connected and that the peroxisomal ABC transporters play a crucial role in de novo ether lipid synthesis.     

Hydroxychloroquine and QTc prolongation in patients with COVID-19: A systematic review and meta-analysis

BackgroundAmong many drugs that hold potential in COVID-19 pandemic, chloroquine (CQ), and its derivative hydroxychloroquine (HCQ) have generated unusual interest. With increasing usage, there has been growing concern about the prolongation of QTc interval and Torsades de Pointes (TdP) with HCQ, especially in combination with azithromycin.AimsThis meta-analysis is planned to study the risk of QTc prolongation and Torsades de pointes (TdP) by a well-defined criterion for HCQ, CQ alone, and in combination with Azithromycin in patients with COVID-19.MethodsA comprehensive literature search was made in two databases (PubMed, Embase). Three outcomes explored in the included studies were frequency of QTc > 500 ms (ms) or ΔQTc > 60 ms (Outcome 1), frequency of QTc > 500 ms (Outcome 2) and frequency of TdP (Outcome 3). Random effects method with inverse variance approach was used for computation of pooled summary and risk ratio.ResultsA total of 13 studies comprising of 2138 patients were included in the final analysis. The pooled prevalence of outcome 1, outcome 2 and outcome 3 for HCQ, CQ with or without Azithromycin were 10.18% (5.59–17.82%, I² – 92%), 10.22% (6.01–16.85%, I² – 79%), and 0.72% (0.34–1.51, I² – 0%) respectively. The prevalence of outcome 2 in subgroup analysis for HCQ and HCQ + Azithromycin was 7.25% (3.22–15.52, I² – 59%) and 8.61% (4.52–15.79, I² – 76%), respectively. The risk ratio (RR) for outcome 1 and outcome 2 between HCQ + Azithromycin and HCQ was 1.22 (0.77–1.93, I² – 0%) & 1.51 (0.79–2.87, I² – 13%), respectively and was not significant. Heterogeneity was noted statistically as well clinically (regimen types, patient numbers, study design, and outcome definition).ConclusionThe use of HCQ/CQ is associated with a high prevalence of QTc prolongation. However, it is not associated with a high risk of TdP.     

A novel deep learning method for maize disease identification based on small sample-size and complex background datasets

Maize diseases are a major source of yield loss, but due to the lack of human experience and limitations of traditional image-recognition technology, obtaining satisfactory large-scale identification results of maize diseases are difficult. Fortunately, the advancement of deep learning-based technology makes it possible to automatically identify diseases. However, it still faces issues caused by small sample sizes and complex field background, which affect the accuracy of disease identification. To address these issues, a deep learning-based method was proposed for maize disease identification in this paper. DenseNet121 was used as the main extraction network and a multi-dilated-CBAM-DenseNet (MDCDenseNet) model was built by combining the multi-dilated module and convolutional block attention module (CBAM) attention mechanism. Five models of MDCDenseNet, DenseNet121, ResNet50, MobileNetV2, and NASNetMobile were compared and tested using three kinds of maize leave images from the PlantVillage dataset and field-collected at Northeast Agricultural University in China. Furthermore, auxiliary classifier generative adversarial network (ACGAN) and transfer learning were used to expand the dataset and pre-train for optimal identification results. When tested on field-collected datasets with a complex background, the MDCDenseNet model outperformed compared to these models with an accuracy of 98.84%. Therefore, it can provide a viable reference for the identification of maize leaf diseases collected from the farmland with a small sample size and complex background.     

MLR-OOD: A Markov Chain Based Likelihood Ratio Method for Out-Of-Distribution Detection of Genomic Sequences

Machine learning or deep learning models have been widely used for taxonomic classification of metagenomic sequences and many studies reported high classification accuracy. Such models are usually trained based on sequences in several training classes in hope of accurately classifying unknown sequences into these classes. However, when deploying the classification models on real testing data sets, sequences that do not belong to any of the training classes may be present and are falsely assigned to one of the training classes with high confidence. Such sequences are referred to as out-of-distribution (OOD) sequences and are ubiquitous in metagenomic studies. To address this problem, we develop a deep generative model-based method, MLR-OOD, that measures the probability of a testing sequencing belonging to OOD by the likelihood ratio of the maximum of the in-distribution (ID) class conditional likelihoods and the Markov chain likelihood of the testing sequence measuring the sequence complexity. We compose three different microbial data sets consisting of bacterial, viral, and plasmid sequences for comprehensively benchmarking OOD detection methods. We show that MLR-OOD achieves the state-of-the-art performance demonstrating the generality of MLR-OOD to various types of microbial data sets. It is also shown that MLR-OOD is robust to the GC content, which is a major confounding effect for OOD detection of genomic sequences. In conclusion, MLR-OOD will greatly reduce false positives caused by OOD sequences in metagenomic sequence classification.     

Principles and Methods in Computational Membrane Protein Design

After decades of progress in computational protein design, the design of proteins folding and functioning in lipid membranes appears today as the next frontier. Some notable successes in the de novo design of simplified model membrane protein systems have helped articulate fundamental principles of protein folding, architecture and interaction in the hydrophobic lipid environment. These principles are reviewed here, together with the computational methods and approaches that were used to identify them. We provide an overview of the methodological innovations in the generation of new protein structures and functions and in the development of membrane-specific energy functions. We highlight the opportunities offered by new machine learning approaches applied to protein design, and by new experimental characterization techniques applied to membrane proteins. Although membrane protein design is in its infancy, it appears more reachable than previously thought.     

Validation of low-dose lung cancer PET-CT protocol and PET image improvement using machine learning

PurposeTo conduct a simplified lesion-detection task of a low-dose (LD) PET-CT protocol for frequent lung screening using 30% of the effective PETCT dose and to investigate the feasibility of increasing clinical value of low-statistics scans using machine learning.MethodsWe acquired 33 SD PET images, of which 13 had actual LD (ALD) PET, and simulated LD (SLD) PET images at seven different count levels from the SD PET scans. We employed image quality transfer (IQT), a machine learning algorithm that performs patch-regression to map parameters from low-quality to high-quality images. At each count level, patches extracted from 23 pairs of SD/SLD PET images were used to train three IQT models – global linear, single tree, and random forest regressions with cubic patch sizes of 3 and 5 voxels. The models were then used to estimate SD images from LD images at each count level for 10 unseen subjects. Lesion-detection task was carried out on matched lesion-present and lesion-absent images.ResultsLD PET-CT protocol yielded lesion detectability with sensitivity of 0.98 and specificity of 1. Random forest algorithm with cubic patch size of 5 allowed further 11.7% reduction in the effective PETCT dose without compromising lesion detectability, but underestimated SUV by 30%.ConclusionLD PET-CT protocol was validated for lesion detection using ALD PET scans. Substantial image quality improvement or additional dose reduction while preserving clinical values can be achieved using machine learning methods though SUV quantification may be biased and adjustment of our research protocol is required for clinical use.     

Blocking Lipid Uptake Pathways Does not Prevent Toxicity in Adipose Triglyceride Lipase (ATGL) Deficiency

Lipid accumulation in nonadipose tissues can cause lipotoxicity, leading to cell death and severe organ dysfunction. Adipose triglyceride lipase (ATGL) deficiency causes human neutral lipid storage disease and leads to cardiomyopathy; ATGL deficiency has no current treatment. One possible approach to alleviate this disorder has been to alter the diet and reduce the supply of dietary lipids and, hence, myocardial lipid uptake. However, in this study, when we supplied cardiac Atgl KO mice a low- or high-fat diet, we found that heart lipid accumulation, heart dysfunction, and death were not altered. We next deleted lipid uptake pathways in the ATGL-deficient mice through the generation of double KO mice also deficient in either cardiac lipoprotein lipase or cluster of differentiation 36, which is involved in an lipoprotein lipase-independent pathway for FA uptake in the heart. We show that neither deletion ameliorated ATGL-deficient heart dysfunction. Similarly, we determined that non-lipid-containing media did not prevent lipid accumulation by cultured myocytes; rather, the cells switched to increased de novo FA synthesis. Thus, we conclude that pathological storage of lipids in ATGL deficiency cannot be corrected by reducing heart lipid uptake.     

Long-Term Follow-up of Graves Orbitopathy After Treatment With Short- or Long-Term Methimazole or Radioactive Iodine

ObjectiveThe aim of this study was to compare long-term outcomes in terms of new onset or worsening of Graves orbitopathy (GO) in patients with Graves disease treated with different therapeutic modalities for hyperthyroidism.MethodsA total of 1163 patients with Graves disease were enrolled in this study; 263 patients were treated with radioiodine and 808 patients received methimazole (MMI) therapy for a median of 18 months, of whom 178 patients continued MMI for a total of 96 months (long-term methimazole [LT-MMI]). The thyroid hormonal status and GO were evaluated regularly for a median of 159 months since enrollment.ResultsThe rates of relapse, euthyroidism, and hypothyroidism at the end of follow-up were as follows: radioiodine treatment group: 16%, 22%, and 62%, respectively; short-term MMI group: 59%, 36%, and 5%, respectively; and LT-MMI group: 18%, 80%, and 2%, respectively. During the first 18 months of therapy, worsening of GO (11.5% vs 5.7%) and de novo development of GO (12.5% vs 9.8%) were significantly more frequent after radioiodine treatment (P <.004). Overall worsening and de novo development of GO from >18 to 234 months occurred in 26 (9.9%) patients in the radioiodine group and 8 (4.5%) patients in the LT-MMI group (P <.037). No case of worsening or new onset of GO was observed in patients treated with LT-MMI from >60 to 234 months of follow-up.ConclusionProgression and development of GO were associated more with radioiodine treatment than with MMI treatment; GO may appear de novo or worsen years after radioiodine treatment but not after LT-MMI therapy.     

Mass spectrometry-based Cellular Thermal Shift Assay (CETSA®) for target deconvolution in phenotypic drug discovery

The recent renewed interest in phenotypic drug discovery has concomitantly put a focus on target deconvolution in order to achieve drug-target identification. Even though there are prescribed therapies whose mode of action is not fully understood, knowledge of the primary target will inevitably facilitate the discovery and translation of efficacy from bench to bedside. Elucidating targets and subsequent pathways engaged will also facilitate safety studies and overall development of novel drug candidates. Today, there are several techniques available for identifying the primary target, many of which rely on mass spectrometry (MS) to identify compound – target protein interactions. The Cellular Thermal Shift Assay (CETSA®) is well suited for identifying target engagement between ligands and their protein targets. Several studies have shown that CETSA combined with MS is a powerful technique that allows unlabeled target deconvolution in complex samples such as intact cells and tissues in addition to cell lysates and other protein suspensions. The applicability of CETSA MS for target deconvolution purposes will be discussed and exemplified in this mini review.     

Towards a more precise therapy in cancer: Exploring epigenetic complexity

A plethora of preclinical evidences suggests that pharmacological targeting of epigenetic dysregulation is a potent strategy to combat human diseases. Nevertheless, the implementation of epidrugs in clinical practice is very scarce and mainly limited to haematological malignancies. In this review, we discuss cutting-edge strategies to foster the chemical design, the biological rationale and the clinical trial development of epidrugs. Specifically, we focus on the development of dual hybrids to exploit multitargeting of key epigenetic molecules deregulated in cancer; the study of epigenetic-synthetic lethality interactions as a mechanism to address loss-of-function mutations, and the combination of epidrugs with other therapies such as immunotherapy to avoid acquired chemoresistance and increase therapy sensitivity. By exploring these challenges, among others, the field of epigenetic chemical biology will increase its potential for clinical benefit, and more effective strategies targeting the aberrant epigenome in cancer are likely to be developed both in haematological and solid tumours.     

Identification and characterization of 3-ketosphinganine reductase activity encoded at the BT_0972 locus in Bacteroides thetaiotaomicron

Bacterial sphingolipid synthesis is important for the fitness of gut commensal bacteria with an implied potential for regulating mammalian host physiology. Multiple steps in bacterial sphingolipid synthesis pathways have been characterized previously, with the first step of de novo sphingolipid synthesis being well conserved between bacteria and eukaryotes. In mammals, the subsequent step of de novo sphingolipid synthesis is catalyzed by 3-ketosphinganine reductase, but the protein responsible for this activity in bacteria has remained elusive. In this study, we analyzed the 3-ketosphinganine reductase activity of several candidate proteins in Bacteroides thetaiotaomicron chosen based on sequence similarity to the yeast 3-ketosphinganine reductase gene. We further developed a metabolomics-based 3-ketosphinganine reductase activity assay, which revealed that a gene at the locus BT_0972 encodes a protein capable of converting 3-ketosphinganine to sphinganine. Taken together, these results provide greater insight into pathways for bacterial sphingolipid synthesis that can aid in future efforts to understand how microbial sphingolipid synthesis modulates host-microbe interactions.     

Machine-learning aided in situ drug sensitivity screening predicts treatment outcomes in ovarian PDX tumors

Long-term treatment outcomes for patients with high grade ovarian cancers have not changed despite innovations in therapies. There is no recommended assay for predicting patient response to second-line therapy, thus clinicians must make treatment decisions based on each individual patient. Patient-derived xenograft (PDX) tumors have been shown to predict drug sensitivity in ovarian cancer patients, but the time frame for intraperitoneal (IP) tumor generation, expansion, and drug screening is beyond that for tumor recurrence and platinum resistance to occur, thus results do not have clinical utility. We describe a drug sensitivity screening assay using a drug delivery microdevice implanted for 24 h in subcutaneous (SQ) ovarian PDX tumors to predict treatment outcomes in matched IP PDX tumors in a clinically relevant time frame. The SQ tumor response to local microdose drug exposure was found to be predictive of the growth of matched IP tumors after multi-week systemic therapy using significantly fewer animals (10 SQ vs 206 IP). Multiplexed immunofluorescence image analysis of phenotypic tumor response combined with a machine learning classifier could predict IP treatment outcomes against three second-line cytotoxic therapies with an average AUC of 0.91.