Structure-based, deep-learning models for protein-ligand binding affinity prediction.

The launch of AlphaFold series has brought deep-learning techniques into the molecular structural science. As another crucial problem, structure-based prediction of protein-ligand binding affinity urgently calls for advanced computational techniques. Is deep learning ready to decode this problem? Here we review mainstream structure-based, deep-learning approaches for this problem, focusing on molecular representations, learning architectures and model interpretability. A model taxonomy has been generated. To compensate for the lack of valid comparisons among those models, we realized and evaluated representatives from a uniform basis, with the advantages and shortcomings discussed. This review will potentially benefit structure-based drug discovery and related areas.

Protein-ligand binding affinity prediction based on profiles of intermolecular contacts.

As a key element in structure-based drug design, binding affinity prediction (BAP) for putative protein-ligand complexes can be efficiently achieved by the incorporation of structural descriptors and machine-learning models. However, developing concise descriptors that will lead to accurate and interpretable BAP remains a difficult problem in this field. Herein, we introduce the profiles of intermolecular contacts (IMCPs) as descriptors for machine-learning-based BAP. IMCPs describe each group of protein-ligand contacts by the count and average distance of the group members, and collaborate closely with classical machine-learning models. Performed on multiple validation sets, IMCP-based models often result in better BAP accuracy than those originating from other similar descriptors. Additionally, IMCPs are simple and concise, and easy to interpret in model training. These descriptors highly conclude the structural information of protein-ligand complexes and can be easily updated with personalized profile features. IMCPs have been implemented in the BAP Toolkit on github ( https://github.com/debbydanwang/BAP).

Structure-based protein-ligand interaction fingerprints for binding affinity prediction.

Binding affinity prediction (BAP) using protein-ligand complex structures is crucial to computer-aided drug design, but remains a challenging problem. To achieve efficient and accurate BAP, machine-learning scoring functions (SFs) based on a wide range of descriptors have been developed. Among those descriptors, protein-ligand interaction fingerprints (IFPs) are competitive due to their simple representations, elaborate profiles of key interactions and easy collaborations with machine-learning algorithms. In this paper, we have adopted a building-block-based taxonomy to review a broad range of IFP models, and compared representative IFP-based SFs in target-specific and generic scoring tasks. Atom-pair-counts-based and substructure-based IFPs show great potential in these tasks.

Genotype-determined EGFR-RTK heterodimerization and its effects on drug resistance in lung Cancer treatment revealed by molecular dynamics simulations.

BACKGROUND: Epidermal growth factor receptor (EGFR) and its signaling pathways play a vital role in pathogenesis of lung cancer. By disturbing EGFR signaling, mutations of EGFR may lead to progression of cancer or the emergence of resistance to EGFR-targeted drugs. RESULTS: We investigated the correlation between EGFR mutations and EGFR-receptor tyrosine kinase (RTK) crosstalk in the signaling network, in order to uncover the drug resistance mechanism induced by EGFR mutations. For several EGFR wild type (WT) or mutated proteins, we measured the EGFR-RTK interactions using several computational methods based on molecular dynamics (MD) simulations, including geometrical characterization of the interfaces and conventional estimation of free energy of binding. Geometrical properties, namely the matching rate of atomic solid angles in the interfaces and center-of-mass distances between interacting atoms, were extracted relying on Alpha Shape modeling. For a couple of RTK partners (c-Met, ErbB2 and IGF-1R), results have shown a looser EGFR-RTK crosstalk for the drug-sensitive EGFR mutant while a tighter crosstalk for the drug-resistant mutant. It guarantees the genotype-determined EGFR-RTK crosstalk, and further proposes a potential drug resistance mechanism by amplified EGFR-RTK crosstalk induced by EGFR mutations. CONCLUSIONS: This study will lead to a deeper understanding of EGFR mutation-induced drug resistance mechanisms and promote the design of innovative drugs.

Proteo-chemometrics interaction fingerprints of protein-ligand complexes predict binding affinity.

MOTIVATION: Reliable predictive models of protein-ligand binding affinity are required in many areas of biomedical research. Accurate prediction based on current descriptors or molecular fingerprints (FPs) remains a challenge. We develop novel interaction FPs (IFPs) to encode protein-ligand interactions and use them to improve the prediction. RESULTS: Proteo-chemometrics IFPs (PrtCmm IFPs) formed by combining extended connectivity fingerprints (ECFPs) with the proteo-chemometrics concept. Combining PrtCmm IFPs with machine-learning models led to efficient scoring models, which were validated on the PDBbind v2019 core set and CSAR-HiQ sets. The PrtCmm IFP Score outperformed several other models in predicting protein-ligand binding affinities. Besides, conventional ECFPs were simplified to generate new IFPs, which provided consistent but faster predictions. The relationship between the base atom properties of ECFPs and the accuracy of predictions was also investigated. AVAILABILITY: PrtCmm IFP has been implemented in the IFP Score Toolkit on github (https://github.com/debbydanwang/IFPscore). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Computationally predicting binding affinity in protein-ligand complexes: free energy-based simulations and machine learning-based scoring functions.

Accurately predicting protein-ligand binding affinities can substantially facilitate the drug discovery process, but it remains as a difficult problem. To tackle the challenge, many computational methods have been proposed. Among these methods, free energy-based simulations and machine learning-based scoring functions can potentially provide accurate predictions. In this paper, we review these two classes of methods, following a number of thermodynamic cycles for the free energy-based simulations and a feature-representation taxonomy for the machine learning-based scoring functions. More recent deep learning-based predictions, where a hierarchy of feature representations are generally extracted, are also reviewed. Strengths and weaknesses of the two classes of methods, coupled with future directions for improvements, are comparatively discussed.

Predicting the impacts of mutations on protein-ligand binding affinity based on molecular dynamics simulations and machine learning methods.

PURPOSE: Mutation-induced variation of protein-ligand binding affinity is the key to many genetic diseases and the emergence of drug resistance, and therefore predicting such mutation impacts is of great importance. In this work, we aim to predict the mutation impacts on protein-ligand binding affinity using efficient structure-based, computational methods. METHODS: Relying on consolidated databases of experimentally determined data we characterize the affinity change upon mutation based on a number of local geometrical features and monitor such feature differences upon mutation during molecular dynamics (MD) simulations. The differences are quantified according to average difference, trajectory-wise distance or time-vary differences. Machine-learning methods are employed to predict the mutation impacts using the resulting conventional or time-series features. Predictions based on estimation of energy and based on investigation of molecular descriptors were conducted as benchmarks. RESULTS: Our method (machine-learning techniques using time-series features) outperformed the benchmark methods, especially in terms of the balanced F1 score. Particularly, deep-learning models led to the best prediction performance with distinct improvements in balanced F1 score and a sustained accuracy. CONCLUSION: Our work highlights the effectiveness of the characterization of affinity change upon mutations. Furthermore, deep-learning techniques are well designed for handling the extracted time-series features. This study can lead to a deeper understanding of mutation-induced diseases and resistance, and further guide the development of innovative drug design.

Characterization of high healthcare utilizer groups using administrative data from an electronic medical record database.

BACKGROUND: High utilizers (HUs) are a small group of patients who impose a disproportionately high burden on the healthcare system due to their elevated resource use. Identification of persistent HUs is pertinent as interventions have not been effective due to regression to the mean in majority of patients. This study will use cost and utilization metrics to segment a hospital-based patient population into HU groups. METHODS: The index visit for each adult patient to an Academic Medical Centre in Singapore during 2006 to 2012 was identified. Cost, length of stay (LOS) and number of specialist outpatient clinic (SOC) visits within 1 year following the index visit were extracted and aggregated. Patients were HUs if they exceeded the 90th percentile of any metric, and Non-HU otherwise. Seven different HU groups and a Non-HU group were constructed. The groups were described in terms of cost and utilization patterns, socio-demographic information, multi-morbidity scores and medical history. Logistic regression compared the groups' persistence as a HU in any group into the subsequent year, adjusting for socio-demographic information and diagnosis history. RESULTS: A total of 388,162 patients above the age of 21 were included in the study. Cost-LOS-SOC HUs had the highest multi-morbidity and persistence into the second year. Common conditions among Cost-LOS and Cost-LOS-SOC HUs were cardiovascular disease, acute cerebrovascular disease and pneumonia, while most LOS and LOS-SOC HUs were diagnosed with at least one mental health condition. Regression analyses revealed that HUs across all groups were more likely to persist compared to Non-HUs, with stronger relationships seen in groups with high SOC utilization. Similar trends remained after further adjustment. CONCLUSION: HUs of healthcare services are a diverse group and can be further segmented into different subgroups based on cost and utilization patterns. Segmentation by these metrics revealed differences in socio-demographic characteristics, disease profile and persistence. Most HUs did not persist in their high utilization, and high SOC users should be prioritized for further longitudinal analyses. Segmentation will enable policy makers to better identify the diverse needs of patients, detect gaps in current care and focus their efforts in delivering care relevant and tailored to each segment.

Drivers of hospital expenditure and length of stay in an academic medical centre: a retrospective cross-sectional study.

BACKGROUND: As healthcare expenditure and utilization continue to rise, understanding key drivers of hospital expenditure and utilization is crucial in policy development and service planning. This study aims to investigate micro drivers of hospital expenditure and length of stay (LOS) in an Academic Medical Centre. METHODS: Data corresponding to 285,767 patients and 207,426 inpatient visits was extracted from electronic medical records of the National University of Hospital in Singapore between 2005 to 2013. Generalized linear models and generalized estimating equations were employed to build patient and inpatient visit models respectively. The patient models provide insight on the factors affecting overall expenditure and LOS, whereas the inpatient visit models provide insight on how expenditure and LOS accumulate longitudinally. RESULTS: Although adjusted expenditure and LOS per inpatient visit were largely similar across socio-economic status (SES) groups, patients of lower SES groups accumulated greater expenditure and LOS over time due to more frequent visits. Admission to a ward class with greater government subsidies was associated with higher expenditure and LOS per inpatient visit. Inpatient death was also associated with higher expenditure per inpatient visit. Conditions that drove patient expenditure and LOS were largely similar, with mental illnesses affecting LOS to a larger extent. These observations on condition drivers largely held true at visit-level. CONCLUSIONS: The findings highlight the importance of distinguishing the drivers of patient expenditure and inpatient utilization at the patient-level from those at the visit-level. This allows better understanding of the drivers of healthcare utilization and how utilization accumulates longitudinally, important for health policy and service planning.

Joint Learning of Multiple Differential Networks With Latent Variables.

Graphical models have been widely used to learn the conditional dependence structures among random variables. In many controlled experiments, such as the studies of disease or drug effectiveness, learning the structural changes of graphical models under two different conditions is of great importance. However, most existing graphical models are developed for estimating a single graph and based on a tacit assumption that there is no missing relevant variables, which wastes the common information provided by multiple heterogeneous data sets and underestimates the influence of latent/unobserved relevant variables. In this paper, we propose a joint differential network analysis (JDNA) model to jointly estimate multiple differential networks with latent variables from multiple data sets. The JDNA model is built on a penalized D-trace loss function, with group lasso or generalized fused lasso penalties. We implement a proximal gradient-based alternating direction method of multipliers to tackle the corresponding convex optimization problems. Extensive simulation experiments demonstrate that JDNA model outperforms state-of-the-art methods in estimating the structural changes of graphical models. Moreover, a series of experiments on several real-world data sets have been performed and experiment results consistently show that our proposed JDNA model is effective in identifying differential networks under different conditions.

Processing of Electronic Medical Records for Health Services Research in an Academic Medical Center: Methods and Validation.

BACKGROUND: Electronic medical records (EMRs) contain a wealth of information that can support data-driven decision making in health care policy design and service planning. Although research using EMRs has become increasingly prevalent, challenges such as coding inconsistency, data validity, and lack of suitable measures in important domains still hinder the progress. OBJECTIVE: The objective of this study was to design a structured way to process records in administrative EMR systems for health services research and assess validity in selected areas. METHODS: On the basis of a local hospital EMR system in Singapore, we developed a structured framework for EMR data processing, including standardization and phenotyping of diagnosis codes, construction of cohort with multilevel views, and generation of variables and proxy measures to supplement primary data. Disease complexity was estimated by Charlson Comorbidity Index (CCI) and Polypharmacy Score (PPS), whereas socioeconomic status (SES) was estimated by housing type. Validity of modified diagnosis codes and derived measures were investigated. RESULTS: Visit-level (N=7,778,761) and patient-level records (n=549,109) were generated. The International Classification of Diseases, Tenth Revision, Australian Modification (ICD-10-AM) codes were standardized to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) with a mapping rate of 87.1%. In all, 97.4% of the ICD-9-CM codes were phenotyped successfully using Clinical Classification Software by Agency for Healthcare Research and Quality. Diagnosis codes that underwent modification (truncation or zero addition) in standardization and phenotyping procedures had the modification validated by physicians, with validity rates of more than 90%. Disease complexity measures (CCI and PPS) and SES were found to be valid and robust after a correlation analysis and a multivariate regression analysis. CCI and PPS were correlated with each other and positively correlated with health care utilization measures. Larger housing type was associated with lower government subsidies received, suggesting association with higher SES. Profile of constructed cohorts showed differences in disease prevalence, disease complexity, and health care utilization in those aged above 65 years and those aged 65 years or younger. CONCLUSIONS: The framework proposed in this study would be useful for other researchers working with EMR data for health services research. Further analyses would be needed to better understand differences observed in the cohorts.

Deciphering mechanisms of acquired T790M mutation after EGFR inhibitors for NSCLC by computational simulations.

Metastatic non-small-cell lung cancer (NSCLC) with activating EGFR mutations responds very well to first and second generation tyrosine-kinase inhibitors (TKI) including gefitinib, erlotinib and afatinib. Unfortunately, drug resistance will eventually develop and about half of the cases are secondary to the emergence of acquired T790M somatic mutation. In this work, we prospectively recruited 68 patients with metastatic EGFR-mutated NSCLC who have developed progressive disease after first-line TKI with or without subsequent TKI and/or other systemic therapy. Liquid biopsy after progression to their last line of systemic therapy were taken for detection of acquired T790M mutation. By performing attribute ranking we found that several attributes, including the initial EGFR mutational type, had a high correlation with the presence of acquired T790M mutation. We also conducted computational studies and discovered that the EGFR mutation delE746_A750 had a lower stability around the residue T790 than delS752_I759 and L858R, which was consistent with our clinical observation that patients with delE746_A750 were more likely to acquire T790M mutation than those with delS752_I759 or L858R. Our results provided new insight to future direction of research on investigating the mechanisms of acquired T790M mutation, which is essential to the development of novel mutation-specific TKIs.

An Eigen-Binding Site Based Method for the Analysis of Anti-EGFR Drug Resistance in Lung Cancer Treatment.

We explore the drug resistance mechanism in non-small cell lung cancer treatment by characterizing the drug-binding site of a protein mutant based on local surface and energy features. These features are transformed to an eigen-binding site space and used for drug resistance level prediction and analysis.

Identifying Multi-Dimensional Co-Clusters in Tensors Based on Hyperplane Detection in Singular Vector Spaces.

Co-clustering, often called biclustering for two-dimensional data, has found many applications, such as gene expression data analysis and text mining. Nowadays, a variety of multi-dimensional arrays (tensors) frequently occur in data analysis tasks, and co-clustering techniques play a key role in dealing with such datasets. Co-clusters represent coherent patterns and exhibit important properties along all the modes. Development of robust co-clustering techniques is important for the detection and analysis of these patterns. In this paper, a co-clustering method based on hyperplane detection in singular vector spaces (HDSVS) is proposed. Specifically in this method, higher-order singular value decomposition (HOSVD) transforms a tensor into a core part and a singular vector matrix along each mode, whose row vectors can be clustered by a linear grouping algorithm (LGA). Meanwhile, hyperplanar patterns are extracted and successfully supported the identification of multi-dimensional co-clusters. To validate HDSVS, a number of synthetic and biological tensors were adopted. The synthetic tensors attested a favorable performance of this algorithm on noisy or overlapped data. Experiments with gene expression data and lineage data of embryonic cells further verified the reliability of HDSVS to practical problems. Moreover, the detected co-clusters are well consistent with important genetic pathways and gene ontology annotations. Finally, a series of comparisons between HDSVS and state-of-the-art methods on synthetic tensors and a yeast gene expression tensor were implemented, verifying the robust and stable performance of our method.

Selectivity profile of afatinib for EGFR-mutated non-small-cell lung cancer.

EGFR-mutated non-small-cell lung cancer (NSCLC) has long been a research focus in lung cancer studies. Besides reversible tyrosine kinase inhibitors (TKIs), new-generation irreversible inhibitors, such as afatinib, embark on playing an important role in NSCLC treatment. To achieve an optimal application of these inhibitors, the correlation between the EGFR mutation status and the potency of such an inhibitor should be decoded. In this study, the correlation was profiled for afatinib, based on a cohort of patients with the EGFR-mutated NSCLC. Relying on extracted DNAs from the paraffin-embedded tumor samples, EGFR mutations were detected by direct sequencing. Progression-free survival (PFS) and the response level were recorded as study endpoints. These PFS and response values were analyzed and correlated to different mutation types, implying a higher potency of afatinib to classic activation mutations (L858R and deletion 19) and a lower one to T790M-related mutations. To further bridge the mutation status with afatinib-related response or PFS, we conducted a computational study to estimate the binding affinity in a mutant-afatinib system, based on molecular structural modeling and dynamics simulations. The derived binding affinities were well in accordance with the clinical response or PFS values. At last, these computational binding affinities were successfully mapped to the patient response or PFS according to linear models. Consequently, a detailed mutation-response or mutation-PFS profile was drafted for afatinib, implying the selective nature of afatinib to various EGFR mutants and further encouraging the design of specialized therapies or innovative drugs.

Contribution of EGFR and ErbB-3 Heterodimerization to the EGFR Mutation-Induced Gefitinib- and Erlotinib-Resistance in Non-Small-Cell Lung Carcinoma Treatments.

EGFR mutation-induced drug resistance has become a major threat to the treatment of non-small-cell lung carcinoma. Essentially, the resistance mechanism involves modifications of the intracellular signaling pathways. In our work, we separately investigated the EGFR and ErbB-3 heterodimerization, regarded as the origin of intracellular signaling pathways. On one hand, we combined the molecular interaction in EGFR heterodimerization with that between the EGFR tyrosine kinase and its inhibitor. For 168 clinical subjects, we characterized their corresponding EGFR mutations using molecular interactions, with three potential dimerization partners (ErbB-2, IGF-1R and c-Met) of EGFR and two of its small molecule inhibitors (gefitinib and erlotinib). Based on molecular dynamics simulations and structural analysis, we modeled these mutant-partner or mutant-inhibitor interactions using binding free energy and its components. As a consequence, the mutant-partner interactions are amplified for mutants L858R and L858R_T790M, compared to the wild type EGFR. Mutant delL747_P753insS represents the largest difference between the mutant-IGF-1R interaction and the mutant-inhibitor interaction, which explains the shorter progression-free survival of an inhibitor to this mutant type. Besides, feature sets including different energy components were constructed, and efficient regression trees were applied to map these features to the progression-free survival of an inhibitor. On the other hand, we comparably examined the interactions between ErbB-3 and its partners (EGFR mutants, IGF-1R, ErbB-2 and c-Met). Compared to others, c-Met shows a remarkably-strong binding with ErbB-3, implying its significant role in regulating ErbB-3 signaling. Moreover, EGFR mutants corresponding to poor clinical outcomes, such as L858R_T790M, possess lower binding affinities with ErbB-3 than c-Met does. This may promote the communication between ErbB-3 and c-Met in these cancer cells. The analysis verified the important contribution of IGF-1R or c-Met in the drug resistance mechanism developed in lung cancer treatments, which may bring many benefits to specialized therapy design and innovative drug discovery.

EGFR Mutant Structural Database: computationally predicted 3D structures and the corresponding binding free energies with gefitinib and erlotinib.

BACKGROUND: Epidermal growth factor receptor (EGFR) mutation-induced drug resistance has caused great difficulties in the treatment of non-small-cell lung cancer (NSCLC). However, structural information is available for just a few EGFR mutants. In this study, we created an EGFR Mutant Structural Database (freely available at http://bcc.ee.cityu.edu.hk/data/EGFR.html ), including the 3D EGFR mutant structures and their corresponding binding free energies with two commonly used inhibitors (gefitinib and erlotinib). RESULTS: We collected the information of 942 NSCLC patients belonging to 112 mutation types. These mutation types are divided into five groups (insertion, deletion, duplication, modification and substitution), and substitution accounts for 61.61% of the mutation types and 54.14% of all the patients. Among all the 942 patients, 388 cases experienced a mutation at residue site 858 with leucine replaced by arginine (L858R), making it the most common mutation type. Moreover, 36 (32.14%) mutation types occur at exon 19, and 419 (44.48%) patients carried a mutation at exon 21. In this study, we predicted the EGFR mutant structures using Rosetta with the collected mutation types. In addition, Amber was employed to refine the structures followed by calculating the binding free energies of mutant-drug complexes. CONCLUSIONS: The EGFR Mutant Structural Database provides resources of 3D structures and the binding affinity with inhibitors, which can be used by other researchers to study NSCLC further and by medical doctors as reference for NSCLC treatment.

Decoding the EGFR mutation-induced drug resistance in lung cancer treatment by local surface geometric properties.

Epidermal growth factor receptor (EGFR) mutation-induced drug resistance leads to a limited efficacy of tyrosine kinase inhibitors during lung cancer treatments. In this study, we explore the correlations between the local surface geometric properties of EGFR mutants and the progression-free survival (PFS). The geometric properties include local surface changes (four types) of the EGFR mutants compared with the wild-type EGFR, and the convex degrees of these local surfaces. Our analysis results show that the Spearman׳s rank correlation coefficients between the PFS and three types of local surface properties are all greater than 0.6 with small P-values, implying a high significance. Moreover, the number of atoms with solid angles in the ranges of [0.71, 1], [0.61, 1] or [0.5, 1], indicating the convex degree of a local EGFR surface, also shows a strong correlation with the PFS. Overall, these characteristics can be efficiently applied to the prediction of drug resistance in lung cancer treatments, and easily extended to other cancer treatments.

Non-Naive Bayesian Classifiers for Classification Problems With Continuous Attributes.

An important way to improve the performance of naive Bayesian classifiers (NBCs) is to remove or relax the fundamental assumption of independence among the attributes, which usually results in an estimation of joint probability density function (p.d.f.) instead of the estimation of marginal p.d.f. in the NBC design. This paper proposes a non-naive Bayesian classifier (NNBC) in which the independence assumption is removed and the marginal p.d.f. estimation is replaced by the joint p.d.f. estimation. A new technique of estimating the class-conditional p.d.f. based on the optimal bandwidth selection, which is the crucial part of the joint p.d.f. estimation, is applied in our NNBC. Three well-known indexes for measuring the performance of Bayesian classifiers, which are classification accuracy, area under receiver operating characteristic curve, and probability mean square error, are adopted to conduct a comparison among the four Bayesian models, i.e., normal naive Bayesian, flexible naive Bayesian (FNB), the homologous model of FNB (FNBROT), and our proposed NNBC. The comparative results show that NNBC is statistically superior to the other three models regarding the three indexes. And, in the comparison with support vector machine and four boosting-based classification methods, NNBC achieves a relatively favorable classification accuracy while significantly reducing the training time.

Transcriptional protein-protein cooperativity in POU/HMG/DNA complexes revealed by normal mode analysis.

Biomolecular cooperativity is of great scientific interest due to its role in biological processes. Two transcription factors (TFs), Oct-4 and Sox-2, are crucial in transcriptional regulation of embryonic stem cells. In this paper, we analyze how Oct-1 (a similar POU factor) and Sox-2, interact cooperatively at their enhancer binding sites in collective motions. Normal mode analysis (NMA) is implemented to study the collective motions of two complexes with each involving these TFs and an enhancer. The special structure of Oct proteins is analyzed comprehensively, after which each Oct/Sox group is reassembled into two protein pairs. We subsequently propose a segmentation idea to extract the most correlated segments in each pair, using correlations of motion magnitude curves. The median analysis on these correlation values shows the intimacy of subunit POUS (Oct-1) and Sox-2. Using those larger-than-median correlation values, we conduct statistical studies and propose several protein-protein cooperative modes (S and D) coupled with their subtypes. Additional filters are applied and similar results are obtained. A supplementary study on the rotation angle curves reaches an agreement with these modes. Overall, these proposed cooperative modes provide useful information for us to understand the complicated interaction mechanism in the POU/HMG/DNA complexes.