Drug Activity Prediction Research Articles

TransCDR: a deep learning model for enhancing the generalizability of drug activity prediction through transfer learning and multimodal data fusion

BackgroundAccurate and robust drug response prediction is of utmost importance in precision medicine. Although many models have been developed to utilize the representations of drugs and cancer cell lines for predicting cancer drug responses (CDR), their performances can be improved by addressing issues such as insufficient data modality, suboptimal fusion algorithms, and poor generalizability for novel drugs or cell lines.ResultsWe introduce TransCDR, which uses transfer learning to learn drug representations and fuses multi-modality features of drugs and cell lines by a self-attention mechanism, to predict the IC50 values or sensitive states of drugs on cell lines. We are the first to systematically evaluate the generalization of the CDR prediction model to novel (i.e., never-before-seen) compound scaffolds and cell line clusters. TransCDR shows better generalizability than 8 state-of-the-art models. TransCDR outperforms its 5 variants that train drug encoders (i.e., RNN and AttentiveFP) from scratch under various scenarios. The most critical contributors among multiple drug notations and omics profiles are Extended Connectivity Fingerprint and genetic mutation. Additionally, the attention-based fusion module further enhances the predictive performance of TransCDR. TransCDR, trained on the GDSC dataset, demonstrates strong predictive performance on the external testing set CCLE. It is also utilized to predict missing CDRs on GDSC. Moreover, we investigate the biological mechanisms underlying drug response by classifying 7675 patients from TCGA into drug-sensitive or drug-resistant groups, followed by a Gene Set Enrichment Analysis.ConclusionsTransCDR emerges as a potent tool with significant potential in drug response prediction.

The immunological landscape and silico analysis of key paraptosis regulator LPAR1 in gastric cancer patients

This study aims to identify key regulators of paraptosis in gastric cancer (GC) and explore their potential in guiding therapeutic strategies, especially in stomach adenocarcinoma (STAD). Genes associated with paraptosis were identified from the references and subjected to Cox regression analysis in the TCGA-STAD cohort. Using machine learning models, LPAR1 consistently ranked highest in feature importance. Multiple sequencing data showed that LPAR1 was significantly overexpressed in cancer-associated fibroblasts (CAFs). LPAR1 expression was significantly higher in normal tissues, and ROC analysis demonstrated its discriminative ability. Copy number alterations and microsatellite instability were significantly associated with LPAR1 expression. High LPAR1 expression correlated with advanced tumor grades and specific cancer immune subtypes, and multivariate analysis confirmed LPAR1 as an independent predictor of poor prognosis. LPAR1 expression was associated with different immune response metrics, including immune effector activation and upregulated chemokine secretion. High LPAR1 expression also correlated with increased sensitivity to compounds, such as BET bromodomain inhibitors I-BET151 and RITA, suggesting LPAR1 as a biomarker for predicting drug activity. FOXP2 showed a strong positive correlation with LPAR1 transcriptional regulation, while increased methylation of LPAR1 promoter regions was negatively correlated with gene expression. Knockdown of LPAR1 affected cell growth in most tumor cell lines, and in vitro experiments demonstrated that LPAR1 influenced extracellular matrix (ECM) contraction and cell viability in the paraptosis of CAFs. These findings suggest that LPAR1 is a critical regulator of paraptosis in GC and a potential biomarker for drug sensitivity and immunotherapy response. This underscores the role of CAFs in mediating tumorigenic effects and suggests that targeting LPAR1 could be a promising strategy for precision medicine in GC.

Developing MLP based prediction system for anticancer drug response using hybrid features of genomics and cheminformatics

Traditional cancer treatment methods have become less effective due to the increasing diversity of cancer types. To address this, precision medicine has gained support within the medical community. This approach tailors treatment to individual patients based on their specific disease characteristics. However, a major challenge lies in accurately predicting how a patient will respond to a specialized drug. Numerous machine learning-based predictive systems have been developed to address this challenge. These systems utilize genomic signatures and the chemical structure of drugs to predict drug activity. In this paper, we introduce a Multi-Layer Perceptron (MLP) based system for predicting the response of anticancer drugs. Our system utilizes hybrid features derived from both genetic expression and the chemical structure of drugs. It is developed using the well-known GDSC dataset (Genomics of Drug Sensitivity in Cancer). Our system achieved a lower Root Mean Square Error (RMSE) value of 0.889, in contrast to the RMSE value of 0.983 obtained by the current state-of-the-art (SOTA) system, SwNet. This indicates superior predictive accuracy. The findings suggest that our proposed research holds promise for the development of targeted drugs for anticancer treatments.

Predicting drug activity against cancer through genomic profiles and SMILES

Due to the constant increase in cancer rates, the disease has become a leading cause of death worldwide, enhancing the need for its detection and treatment. In the era of personalized medicine, the main goal is to incorporate individual variability in order to choose more precisely which therapy and prevention strategies suit each person. However, predicting the sensitivity of tumors to anticancer treatments remains a challenge. In this work, we propose two deep neural network models to predict the impact of anticancer drugs in tumors through the half-maximal inhibitory concentration (IC50). These models join biological and chemical data to apprehend relevant features of the genetic profile and the drug compounds, respectively. In order to predict the drug response in cancer cell lines, this study employed different DL methods, resorting to Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs). In the first stage, two autoencoders were pre-trained with high-dimensional gene expression and mutation data of tumors. Afterward, this genetic background is transferred to the prediction models that return the IC50 value that portrays the potency of a substance in inhibiting a cancer cell line. When comparing RSEM Expected counts and TPM as methods for displaying gene expression data, RSEM has been shown to perform better in deep models and CNNs model can obtain better insight in these types of data. Moreover, the obtained results reflect the effectiveness of the extracted deep representations in the prediction of the IC50 value that portrays the potency of a substance in inhibiting a tumor, achieving a performance of a mean squared error of 1.06 and surpassing previous state-of-the-art models.

Biological Activity of Late Transition Metal‐Based Compounds: From Computational and Theoretical Studies to Laboratory Exploration and Beyond

The use of metal compounds such as cisplatin and its derivatives as therapeutic and diagnostic agents against various diseases is well established. Although metallodrugs have been very successful clinically, low therapeutic selectivity and the potential toxicity of the metal compounds mandate the need for a continuous search for more efficient and specific treatments. Hence, the use of computer‐aided drug design (CADD), molecular modelling and theoretical studies in designing, selecting and applying potential drug candidates have become ever more necessary. Among the many computational and theoretical techniques, molecular docking and density functional theory play significant roles in predicting drug activity. Recent methodologies, within the past 5 years, that rely on these techniques to advance the adoption of potential metallodrugs are highlighted and thoroughly discussed. This is because advancements in computing power have led to the wide use of CADD approaches for drug discovery, development and analysis to shorten the time and reduce the costs associated with drug discovery.

Chemical complexity challenge: Is multi‐instance machine learning a solution?

AbstractMolecules are complex dynamic objects that can exist in different molecular forms (conformations, tautomers, stereoisomers, protonation states, etc.) and often it is not known which molecular form is responsible for observed physicochemical and biological properties of a given molecule. This raises the problem of the selection of the correct molecular form for machine learning modeling of target properties. The same problem is common to biological molecules (RNA, DNA, proteins)—long sequences where only key segments, which often cannot be located precisely, are involved in biological functions. Multi‐instance machine learning (MIL) is an efficient approach for solving problems where objects under study cannot be uniquely represented by a single instance, but rather by a set of multiple alternative instances. Multi‐instance learning was formalized in 1997 and motivated by the problem of conformation selection in drug activity prediction tasks. Since then MIL has found a lot of applications in various domains, such as information retrieval, computer vision, signal processing, bankruptcy prediction, and so on. In the given review we describe the MIL framework and its applications to the tasks associated with ambiguity in the representation of small and biological molecules in chemoinformatics and bioinformatics. We have collected examples that demonstrate the advantages of MIL over the traditional single‐instance learning (SIL) approach. Special attention was paid to the ability of MIL models to identify key instances responsible for a modeling property.This article is categorized under: Data Science > Chemoinformatics Data Science > Artificial Intelligence/Machine Learning

A Comprehensive Review on Multiple Instance Learning

Multiple-instance learning has become popular over recent years due to its use in some special scenarios. It is basically a type of weakly supervised learning where the learning dataset contains bags of instances instead of a single feature vector. Each bag is associated with a single label. This type of learning is flexible and a natural fit for multiple real-world problems. MIL has been employed to deal with a number of challenges, including object detection and identification tasks, content-based image retrieval, and computer-aided diagnosis. Medical image analysis and drug activity prediction have been the main uses of MIL in biomedical research. Many Algorithms based on MIL have been put forth over the years. In this paper, we will discuss MIL, the background of MIL and its application in multiple domains, some MIL-based methods, challenges, and lastly, the conclusions and prospects.

Affinity character analysis of magnolol and honokiol based on stepwise frontal analysis coupled with cell membrane chromatography

Magnolol and honokiol have been reported to exhibit anti-cancer activity. However, few studies are in relation to the interaction of magnolol/honokiol with vascular endothelial growth factor 2 (VEGFR2). In this study, a membrane chromatography method based on VEGFR2 was established for the interaction characteristic analysis between drug and receptor. The selectivity, repeatability and stability of the chromatographic model were evaluated using drugs acting on different receptors. The affinity between VEGFR2 and magnolol/honokiol was verified by cell membrane chromatography. The binding sites of magnolol/honokiol and VEGFR2 were analyzed by zonal elution. Especially, the dissociation equilibrium constants (Kd) of magnolol/honokiol and VEGFR2 were measured by zonal elution and stepwise frontal analysis respectively. In addition, the actions of magnolol/honokiol on VEGFR2 were analyzed by stepwise frontal analysis at different temperatures. The results showed that the binding sites of magnolol and honokiol on VEGFR2 were different from sorafenib, indicating that magnolol and honokiol could be used as competitive agents for self-competitive displacement experiment. The Kd values (order of magnitude) of magnolol/honokiol with VEGFR2 measured by stepwise frontal analysis were consistent with the zonal elution results. Honokiol binds VEGFR2 with higher affinity than magnolol. The main forces that stabilize the interactions of honokiol with VEGFR2 are hydrogen bonds and van der Waal’s forces, and the main force of magnolol is electrostatic forces. These discoveries could assist in the prediction of drug activity and understanding for the underlying mechanism.

Molecular machine learning with conformer ensembles

Virtual screening can accelerate drug discovery by identifying promising candidates for experimental evaluation. Machine learning is a powerful method for screening, as it can learn complex structure–property relationships from experimental data and make rapid predictions over virtual libraries. Molecules inherently exist as a three-dimensional ensemble and their biological action typically occurs through supramolecular recognition. However, most deep learning approaches to molecular property prediction use a 2D graph representation as input, and in some cases a single 3D conformation. Here we investigate how the 3D information of multiple conformers, traditionally known as 4D information in the cheminformatics community, can improve molecular property prediction in deep learning models. We introduce multiple deep learning models that expand upon key architectures such as ChemProp and SchNet, adding elements such as multiple-conformer inputs and conformer attention. We then benchmark the performance trade-offs of these models on 2D, 3D and 4D representations in the prediction of drug activity using a large training set of geometrically resolved molecules. The new architectures perform significantly better than 2D models, but their performance is often just as strong with a single conformer as with many. We also find that 4D deep learning models learn interpretable attention weights for each conformer.

Application of three-dimensional cell culture technology in screening anticancer drugs.

The drug development process involves a variety of drug activity evaluations, which can determine drug efficacy, strictly analyze the biological indicators after the drug action, and use these indicators as the preclinical drug evaluation criteria. At present, most of the screening of preclinical anticancer drugs mainly relies on traditional 2D cell culture. However, this traditional technology cannot simulate the tumor microenvironment in vivo, let alone reflect the characteristics of solid tumors in vivo, and has a relatively poor ability to predict drug activity. 3D cell culture is a technology between 2D cell culture and animal experiments, which can better reflect the biological state in vivo and reduce the consumption of animal experiments. 3D cell culture can link the individual study of cells with the study of the whole organism, reproduce in vitro the biological phenotype of cells in vivo more greatly, and thus predict the activity and resistance of anti-tumor drugs more accurately. In this paper, the common techniques of 3D cell culture are discussed, with emphasis on its main advantages and application in the evaluation of anti-tumor resistance, which can provide strategies for the screening of anti-tumor drugs.

MOViDA: multiomics visible drug activity prediction with a biologically informed neural network model.

The process of drug development is inherently complex, marked by extended intervals from the inception of a pharmaceutical agent to its eventual launch in the market. Additionally, each phase in this process is associated with a significant failure rate, amplifying the inherent challenges of this task. Computational virtual screening powered by machine learning algorithms has emerged as a promising approach for predicting therapeutic efficacy. However, the complex relationships between the features learned by these algorithms can be challenging to decipher. We have engineered an artificial neural network model designed specifically for predicting drug sensitivity. This model utilizes a biologically informed visible neural network, thereby enhancing its interpretability. The trained model allows for an in-depth exploration of the biological pathways integral to prediction and the chemical attributes of drugs that impact sensitivity. Our model harnesses multiomics data derived from a different tumor tissue sources, as well as molecular descriptors that encapsulate the properties of drugs. We extended the model to predict drug synergy, resulting in favorable outcomes while retaining interpretability. Given the imbalanced nature of publicly available drug screening datasets, our model demonstrated superior performance to state-of-the-art visible machine learning algorithms. MOViDA is implemented in Python using PyTorch library and freely available for download at https://github.com/Luigi-Ferraro/MOViDA. Training data, RIS score and drug features are archived on Zenodo https://doi.org/10.5281/zenodo.8180380.

2D, 3D-QSAR study and docking of vascular endothelial growth factor receptor 3 (VEGFR3) inhibitors for potential treatment of retinoblastoma.

Background: Retinoblastoma is currently the most common malignant tumor seen in newborns and children's eyes worldwide, posing a life-threatening hazard. Chemotherapy is an integral part of retinoblastoma treatment. However, the chemotherapeutic agents used in clinics often lead to drug resistance. Thus there is a need to investigate new chemotherapy-targeted agents. VEGFR3 inhibitors are anti-tumour-growth and could be used to develop novel retinoblastoma-targeted agents. Objective: To predict drug activity, discover influencing factors and design new drugs by building 2D, 3D-QSAR models. Method: First, linear and non-linear QSAR models were built using heuristic methods and gene expression programming (GEP). The comparative molecular similarity indices analysis (COMISA) was then used to construct 3D-QSAR models through the SYBYL software. New drugs were designed by changing drug activity factors in both models, and molecular docking experiments were performed. Result: The best linear model created using HM had an R2, S2, and R2cv of 0.82, 0.02, and 0.77, respectively. For the training and test sets, the best non-linear model created using GEP had correlation coefficients of 0.83 and 0.72 with mean errors of 0.02 and 0.04. The 3D model designed using SYBYL passed external validation due to its high Q2 (0.503), R2 (0.805), and F-value (76.52), as well as its low standard error of SEE value (0.172). This demonstrates the model's reliability and excellent predictive ability. Based on the molecular descriptors of the 2D model and the contour plots of the 3D model, we designed 100 new compounds using the best active compound 14 as a template. We performed activity prediction and molecular docking experiments on them, in which compound 14.d performed best regarding combined drug activity and docking ability. Conclusion: The non-linear model created using GEP was more stable and had a more substantial predictive power than the linear model built using the heuristic technique (HM). The compound 14.d designed in this experiment has the potential for anti-retinoblastoma treatment, which provides new design ideas and directions for retinoblastoma-targeted drugs.

In Silico Drug Activity Prediction of Chemical Components of Justicia adhatoda

In Silico Drug Activity Prediction of Chemical Components of Justicia adhatoda

In vitro construction of lung cancer organoids by 3D bioprinting for drug evaluation

The construction of tumor model plays a crucial role in preclinical drug evaluation. However, traditional two-dimensional (2D) monolayer cell models are unable to accurately predict drug activity in vivo, and animal models have species differences and ethical issues. Herein, a droplet-based 3D bioprinting strategy is established to construct lung cancer organoid arrays for drug evaluation. The bioink system is based on sodium alginate (SA, used to crosslink with Ca2+), hyaluronic acid (HA, used to improve gel viscoelasticity and printability), and arginine-glycine-aspartic acid peptide (RGD, used to improve cell adhesion). Hundreds of organoids with 3D multicellular spherical structures are produced in batches after post-printing culture. Compared with 2D monolayer culture models, 3D organoids show higher cell activity and functional expression of P-CK, ProSP-C, MUC1 and Caveolin-1, as well as higher anticancer drug resistance and IC50. This strategy will provide insights into the field of organoid modeling, high-throughput and personalized drug screening.

Discovery of ANO1 Inhibitors based on Machine learning and molecule docking simulation approaches

Calcium-activated chloride channels (CaCCs) are chloride channels that are regulated according to intracellular calcium ion concentrations. The channel protein ANO1 is widely present in cells and is involved in physiological activities including cellular secretion, signaling, cell proliferation and vasoconstriction and diastole. In this study, the ANO1 inhibitors were investigated with machine learning and molecular simulation. Two-dimensional structure-activity relationship (2D-SAR) and three-dimensional quantitative structure-activity relationship (3D-QSAR) models were developed for the qualitative and quantitative prediction of ANO1 inhibitors. The results showed that the prediction accuracies of the model were 85.9% and 87.8% for the training and test sets, respectively, and 85.9% and 87.8% for the rotating forest (RF) in the 2D-SAR model. The CoMFA and CoMSIA methods were then used for 3D QSAR modeling of ANO1 inhibitors, respectively. The q2 coefficients for model cross-validation were all greater than 0.5, implying that we were able to obtain a stable model for drug activity prediction. Molecular docking was further used to simulate the interactions between the five most promising compounds predicted by the model and the ANO1 protein. The total score for the docking results between all five compounds and the target protein was greater than 6, indicating that they interacted strongly in the form of hydrogen bonds. Finally, simulations of amino acid mutations around the docking cavity of the target proteins showed that each molecule had two or more sites of reduced affinity following a single mutation, indicating outstanding specificity of the screened drug molecules and their protein ligands.

A feature transferring workflow between data-poor compounds in various tasks.

Compound screening by in silico approaches has advantages in identifying high-activity leading compounds and can predict the safety of the drug. A key challenge is that the number of observations of drug activity and toxicity accumulation varies by target in different datasets, some of which are more understudied than others. Owing to an overall insufficiency and imbalance of drug data, it is hard to accurately predict drug activity and toxicity of multiple tasks by the existing models. To solve this problem, this paper proposed a two-stage transfer learning workflow to develop a novel prediction model, which can accurately predict drug activity and toxicity of the targets with insufficient observations. We built a balanced dataset based on the Tox21 dataset and developed a drug activity and toxicity prediction model based on Siamese networks and graph convolution to produce multitasking output. We also took advantage of transfer learning from data-rich targets to data-poor targets. We showed greater accuracy in predicting the activity and toxicity of compounds to targets with rich data and poor data. In Tox21, a relatively rich dataset, the prediction model accuracy for classification tasks was 0.877 AUROC. In the other five unbalanced datasets, we also found that transfer learning strategies brought the accuracy of models to a higher level in understudied targets. Our models can overcome the imbalance in target data and predict the compound activity and toxicity of understudied targets to help prioritize upcoming biological experiments.

Class imbalance learning with Bayesian optimization applied in drug discovery

Machine intelligence (MI), including machine learning and deep learning, have been regarded as promising methods to reduce the prohibitively high cost of drug development. However, a dilemma within MI has limited its wide application: machine learning models are easier to interpret but yield worse predictive performance than deep learning models. Therefore, we propose a pipeline called Class Imbalance Learning with Bayesian Optimization (CILBO) to improve the performance of machine learning models in drug discovery. To demonstrate the efficacy of the CILBO pipeline, we developed an example model to predict antibacterial candidates. Comparison of the antibacterial prediction performance between our model and a well-known deep learning model published by Stokes et al. suggests that our model can perform as well as the deep learning model in drug activity prediction. The CILBO pipeline we propose provides a simple, alternative approach to accelerate preliminary screenings and decrease the cost of drug discovery.

P-203 Relationship of myelotoxicity and of gastrointestinal toxicity with the treatment outcome in trials of second-line chemotherapy of patients with metastatic colorectal cancer

Although some toxicities of molecularly targeted therapies have been shown to predict drug activity, this is unclear for traditional cytotoxic chemotherapy. Various studies, mostly in later lines of metastatic disease, have documented that neutropenia after chemotherapy has a predictive role for the efficacy of chemotherapy in patients with mCRC. In particular, it seems that mild neutropenia is the one that is most often associated with the best outcome. After systematic review and selection of the randomized phase III trials of second-line chemotherapy of patients with mCRC, all those that reported bone marrow (neutropenia, anemia, thrombocytopenia) and gastrointestinal (diarrhea, vomiting) toxicities, mild or severe, were included in the analysis. For each trial, the relationship between the difference in the frequency rates of each of the toxicities between study arms with the difference in PFS was evaluated by Pearson's test (rho), in order to detect a possible correlation between toxicity and outcome. Fourteen studies were selected. The difference in neutropenia rates between the study arms correlated with the difference in PFS (rho = 0.817; p-value 0.004; 10 studies). In particular, the correlation was significant for mild neutropenia (rho = 0.764; p-value 0.004; 12 studies). Similar data, though from a smaller number of studies, were detectable for the other bone marrow toxicities but not for the gastrointestinal ones. Mild to moderate bone marrow toxicity could be a predictor of the efficacy of second-line cytotoxic chemotherapy in patients with mCRC, and therefore it could reflect not only the direct toxicity of the drugs, but also a chemotherapy-related response mechanism.

Machine Learning Prediction of Allosteric Drug Activity from Molecular Dynamics.

Allosteric drugs have been attracting increasing interest over the past few years. In this context, it is common practice to use high-throughput screening for the discovery of non-natural allosteric drugs. While the discovery stage is supported by a growing amount of biological information and increasing computing power, major challenges still remain in selecting allosteric ligands and predicting their effect on the target protein’s function. Indeed, allosteric compounds can act both as inhibitors and activators of biological responses. Computational approaches to the problem have focused on variations on the theme of molecular docking coupled to molecular dynamics with the aim of recovering information on the (long-range) modulation typical of allosteric proteins.

Ex Vivo Analysis of Primary Tumor Specimens for Evaluation of Cancer Therapeutics.

The use of ex vivo drug sensitivity testing to predict drug activity in individual patients has been actively explored for almost 50 years without delivering a generally useful predictive capability. However, extended failure should not be an indicator of futility. This is especially true in cancer research where ultimate success is often preceded by less successful attempts. For example, both immune- and genetic-based targeted therapies for cancer underwent numerous failed attempts before biological understanding, improved targets, and optimized drug development matured to facilitate an arsenal of transformational drugs. Similarly, the concept of directly assessing drug sensitivity of primary tumor biopsies-and the use of this information to help direct therapeutic approaches-has a long history with a definitive learning curve. In this review, we will survey the history of ex vivo testing as well as the current state of the art for this field. We will present an update on methodologies and approaches, describe the use of these technologies to test cutting-edge drug classes, and describe an increasingly nuanced understanding of tumor types and models for which this strategy is most likely to succeed. We will consider the relative strengths and weaknesses of predicting drug activity across the broad biological context of cancer patients and tumor types. This will include an analysis of the potential for ex vivo drug sensitivity testing to accurately predict drug activity within each of the biological hallmarks of cancer pathogenesis.