Multi-label classification with XGBoost for metabolic pathway prediction.

BACKGROUND: Metabolic pathway prediction is one possible approach to address the problem in system biology of reconstructing an organism's metabolic network from its genome sequence. Recently there have been developments in machine learning-based pathway prediction methods that conclude that machine learning-based approaches are similar in performance to the most used method, PathoLogic which is a rule-based method. One issue is that previous studies evaluated PathoLogic without taxonomic pruning which decreases its performance. RESULTS: In this study, we update the evaluation results from previous studies to demonstrate that PathoLogic with taxonomic pruning outperforms previous machine learning-based approaches and that further improvements in performance need to be made for them to be competitive. Furthermore, we introduce mlXGPR, a XGBoost-based metabolic pathway prediction method based on the multi-label classification pathway prediction framework introduced from mlLGPR. We also improve on this multi-label framework by utilizing correlations between labels using classifier chains. We propose a ranking method that determines the order of the chain so that lower performing classifiers are placed later in the chain to utilize the correlations between labels more. We evaluate mlXGPR with and without classifier chains on single-organism and multi-organism benchmarks. Our results indicate that mlXGPR outperform other previous pathway prediction methods including PathoLogic with taxonomic pruning in terms of hamming loss, precision and F1 score on single organism benchmarks. CONCLUSIONS: The results from our study indicate that the performance of machine learning-based pathway prediction methods can be substantially improved and can even outperform PathoLogic with taxonomic pruning.

Deep-Learning Algorithm and Concomitant Biomarker Identification for NSCLC Prediction Using Multi-Omics Data Integration.

Early diagnosis of lung cancer to increase the survival rate, which is currently at a low range of mid-30%, remains a critical need. Despite this, multi-omics data have rarely been applied to non-small-cell lung cancer (NSCLC) diagnosis. We developed a multi-omics data-affinitive artificial intelligence algorithm based on the graph convolutional network that integrates mRNA expression, DNA methylation, and DNA sequencing data. This NSCLC prediction model achieved a 93.7% macro F1-score, indicating that values for false positives and negatives were substantially low, which is desirable for accurate classification. Gene ontology enrichment and pathway analysis of features revealed that two major subtypes of NSCLC, lung adenocarcinoma and lung squamous cell carcinoma, have both specific and common GO biological processes. Numerous biomarkers (i.e., microRNA, long non-coding RNA, differentially methylated regions) were newly identified, whereas some biomarkers were consistent with previous findings in NSCLC (e.g., SPRR1B). Thus, using multi-omics data integration, we developed a promising cancer prediction algorithm.

Gene-centric multi-omics integration with convolutional encoders for cancer drug response prediction.

MOTIVATION: Tumor heterogeneity, including genetic and transcriptomic characteristics, can reduce the efficacy of anticancer pharmacological therapy, resulting in clinical variability in patient response to therapeutic medications. Multi-omics integration can allow in silico models to provide an additional perspective on a biological system. METHODS: In this study, we propose a gene-centric multi-channel (GCMC) architecture to integrate multi-omics for predicting cancer drug response. GCMC transformed multi-omics profiles into a three-dimensional tensor with an additional dimension for omics types. GCMC's convolutional encoders captures multi-omics profiles for each gene and yields gene-centric features to predict drug responses. RESULTS: We evaluated GCMC on various datasets, including The Cancer Genome Atlas (TCGA) patients, patient-derived xenografts (PDX) mice models, and the Genomics of Drug Sensitivity in Cancer (GDSC) cell line datasets. GCMC achieved better performance than baseline models, including single-omics models, in more than 75% of 265 drugs from GDSC cell line datasets. Furthermore, as for the clinical applicability of GCMC, it achieved the best performance on TCGA and PDX datasets in terms of both AUPR and AUC. We also analyzed models' capability of integrating multi-omics profiles by measuring the contribution ratio of omics types. GCMC can incorporate multi-omics profiles in various manners to enhance performance for each drug type. These results suggested that GCMC can improve performance and feature extraction capability by integrating multi-omics profiles in a gene-centric manner.

Efficient Wear Simulation Methodology for Predicting Nonlinear Wear Behavior of Tools in Sheet Metal Forming.

In conventional wear simulation, the geometry must be updated for succeeding iterations to predict the accumulated wear. However, repeating this procedure up to the desired iteration is rather time consuming. Thus, a wear simulation process capable of reasonable quantitative wear prediction in reduced computational time is needed. This study aimed to develop an efficient wear simulation method to predict quantitative wear reasonably in reduced computational time without updating the geometry for succeeding iterations. The wear resistance of a stamping tool was quantitatively evaluated for different punch shapes (R3.0 and R5.5) and coating conditions (physical vapor deposition of CrN and AlTiCrN coatings) by using a progressive die set. To capture the nonlinear wear behavior with respect to strokes, a nonlinear equation from a modified form of Archard's wear model was proposed. By utilizing the scale factor representing the changes in wear properties with respect to wear depth as input, the simulation can predict the behavior of rapidly increasing wear depth with respect to strokes after failure initiation. Furthermore, the proposed simulation method is efficient in terms of computational time because it does not need to perform geometry updates.

A novel approach to predicting the synergy of anti-cancer drug combinations using document-based feature extraction.

BACKGROUND: To reduce drug side effects and enhance their therapeutic effect compared with single drugs, drug combination research, combining two or more drugs, is highly important. Conducting in-vivo and in-vitro experiments on a vast number of drug combinations incurs astronomical time and cost. To reduce the number of combinations, researchers classify whether drug combinations are synergistic through in-silico methods. Since unstructured data, such as biomedical documents, include experimental types, methods, and results, it can be beneficial extracting features from documents to predict anti-cancer drug combination synergy. However, few studies predict anti-cancer drug combination synergy using document-extracted features. RESULTS: We present a novel approach for anti-cancer drug combination synergy prediction using document-based feature extraction. Our approach is divided into two steps. First, we extracted documents containing validated anti-cancer drug combinations and cell lines. Drug and cell line synonyms in the extracted documents were converted into representative words, and the documents were preprocessed by tokenization, lemmatization, and stopword removal. Second, the drug and cell line features were extracted from the preprocessed documents, and training data were constructed by feature concatenation. A prediction model based on deep and machine learning was created using the training data. The use of our features yielded higher results compared to the majority of published studies. CONCLUSIONS: Using our prediction model, researchers can save time and cost on new anti-cancer drug combination discoveries. Additionally, since our feature extraction method does not require structuring of unstructured data, new data can be immediately applied without any data scalability issues.

ROS-induced PADI2 downregulation accelerates cellular senescence via the stimulation of SASP production and NFκB activation.

Cellular senescence is closely related to tissue aging including bone. Bone homeostasis is maintained by the tight balance between bone-forming osteoblasts and bone-resorbing osteoclasts, but it undergoes deregulation with age, causing age-associated osteoporosis, a main cause of which is osteoblast dysfunction. Oxidative stress caused by the accumulation of reactive oxygen species (ROS) in bone tissues with aging can accelerate osteoblast senescence and dysfunction. However, the regulatory mechanism that controls the ROS-induced senescence of osteoblasts is poorly understood. Here, we identified Peptidyl arginine deiminase 2 (PADI2), a post-translational modifying enzyme, as a regulator of ROS-accelerated senescence of osteoblasts via RNA-sequencing and further functional validations. PADI2 downregulation by treatment with H2O2 or its siRNA promoted cellular senescence and suppressed osteoblast differentiation. CCL2, 5, and 7 known as the elements of the senescence-associated secretory phenotype (SASP) which is a secretome including proinflammatory cytokines and chemokines emitted by senescent cells and a representative feature of senescence, were upregulated by H2O2 treatment or Padi2 knockdown. Furthermore, blocking these SASP factors with neutralizing antibodies or siRNAs alleviated the senescence and dysfunction of osteoblasts induced by H2O2 treatment or Padi2 knockdown. The elevated production of these SASP factors was mediated by the activation of NFκB signaling pathway. The inhibition of NFκB using the pharmacological inhibitor or siRNA effectively relieved H2O2 treatment- or Padi2 knockdown-induced senescence and osteoblast dysfunction. Together, our study for the first time uncover the role of PADI2 in ROS-accelerated cellular senescence of osteoblasts and provide new mechanistic and therapeutic insights into excessive ROS-promoted cellular senescence and aging-related bone diseases.

COOBoostR: An Extreme Gradient Boosting-Based Tool for Robust Tissue or Cell-of-Origin Prediction of Tumors.

We present here COOBoostR, a computational method designed for the putative prediction of the tissue- or cell-of-origin of various cancer types. COOBoostR leverages regional somatic mutation density information and chromatin mark features to be applied to an extreme gradient boosting-based machine-learning algorithm. COOBoostR ranks chromatin marks from various tissue and cell types, which best explain the somatic mutation density landscape of any sample of interest. A specific tissue or cell type matching the chromatin mark feature with highest explanatory power is designated as a potential tissue- or cell-of-origin. Through integrating either ChIP-seq based chromatin data, along with regional somatic mutation density data derived from normal cells/tissue, precancerous lesions, and cancer types, we show that COOBoostR outperforms existing random forest-based methods in prediction speed, with comparable or better tissue or cell-of-origin prediction performance (prediction accuracy-normal cells/tissue: 76.99%, precancerous lesions: 95.65%, cancer cells: 89.39%). In addition, our results suggest a dynamic somatic mutation accumulation at the normal tissue or cell stage which could be intertwined with the changes in open chromatin marks and enhancer sites. These results further represent chromatin marks shaping the somatic mutation landscape at the early stage of mutation accumulation, possibly even before the initiation of precancerous lesions or neoplasia.

Differences in the heritability of craniofacial skeletal and dental characteristics between twin pairs with skeletal Class I and II malocclusions.

OBJECTIVE: To investigate differences in the heritability of skeletodental characteristics between twin pairs with skeletal Class I and Class II malocclusions. METHODS: Forty Korean adult twin pairs were divided into Class I (C-I) group (0° ≤ angle between point A, nasion, and point B [ANB]) ≤ 4°; mean age, 40.7 years) and Class II (C-II) group (ANB > 4°; mean age, 43.0 years). Each group comprised 14 monozygotic and 6 dizygotic twin pairs. Thirty-three cephalometric variables were measured using lateral cephalograms and were categorized as the anteroposterior, vertical, dental, mandible, and cranial base characteristics. The ACE model was used to calculate heritability (A > 0.7, high heritability). Thereafter, principal component analysis (PCA) was performed. RESULTS: Twin pairs in C-I group exhibited high heritability values in the facial anteroposterior characteristics, inclination of the maxillary and mandibular incisors, mandibular body length, and cranial base angles. Twin pairs in C-II group showed high heritability values in vertical facial height, ramus height, effective mandibular length, and cranial base length. PCA extracted eight components with 88.3% in the C-I group and seven components with 91.0% cumulative explanation in the C-II group. CONCLUSIONS: Differences in the heritability of skeletodental characteristics between twin pairs with skeletal Class I and II malocclusions might provide valuable information for growth prediction and treatment planning.

Prediction of treatment outcome in burning mouth syndrome patients using machine learning based on clinical data.

The purpose of this study is to apply a machine learning approach to predict whether patients with burning mouth syndrome (BMS) respond to the initial approach and clonazepam therapy based on clinical data. Among the patients with the primary type of BMS who visited the clinic from 2006 to 2015, those treated with the initial approach of detailed explanation regarding home care instruction and use of oral topical lubricants, or who were prescribed clonazepam for a minimum of 1 month were included in this study. The clinical data and treatment outcomes were collected from medical records. Extreme Gradient-Boosted Decision Trees was used for machine learning algorithms to construct prediction models. Accuracy of the prediction models was evaluated and feature importance calculated. The accuracy of the prediction models for the initial approach and clonazepam therapy was 67.6% and 67.4%, respectively. Aggravating factors and psychological distress were important features in the prediction model for the initial approach, and intensity of symptoms before administration was the important feature in the prediction model for clonazepam therapy. In conclusion, the analysis of treatment outcomes in patients with BMS using a machine learning approach showed meaningful results of clinical applicability.

Biomedical Entity Explorer: A Web Server for Biomedical Entity Exploration.

Biomedical Entity Explorer (BEE) is a web server that can search for biomedical entities from a database of six biomedical entity types (gene, miRNA, drug, disease, single nucleotide polymorphism [SNP], pathway) and their gene associations. The search results can be explored using intersections, unions, and negations. BEE has integrated biomedical entities from 16 databases (Ensemble, PharmGKB, Genetic Home Reference, Tarbase, Mirbase, NCI Thesaurus, DisGeNET, Linked life data, UMLS, GSEA MsigDB, Reactome, KEGG, Gene Ontology, HGVD, SNPedia, and dbSNP) based on their gene associations and built a database with their synonyms, descriptions, and links containing individual details. Users can enter the keyword of one or more entities and select the type of entity for which they want to know the relationship for and by using set operations such as union, negation, and intersection, they can navigate the search results more clearly. We believe that BEE will not only be useful for biologists querying for complex associations between entities, but can also be a good starting point for general users searching for biomedical entities. BEE is accessible at (http://bike-bee.snu.ac.kr).

Current advances of epigenetics in periodontology from ENCODE project: a review and future perspectives.

BACKGROUND: The Encyclopedia of DNA Elements (ENCODE) project has advanced our knowledge of the functional elements in the genome and epigenome. The aim of this article was to provide the comprehension about current research trends from ENCODE project and establish the link between epigenetics and periodontal diseases based on epigenome studies and seek the future direction. MAIN BODY: Global epigenome research projects have emphasized the importance of epigenetic research for understanding human health and disease, and current international consortia show an improved interest in the importance of oral health with systemic health. The epigenetic studies in dental field have been mainly conducted in periodontology and have focused on DNA methylation analysis. Advances in sequencing technology have broadened the target for epigenetic studies from specific genes to genome-wide analyses. CONCLUSIONS: In line with global research trends, further extended and advanced epigenetic studies would provide crucial information for the realization of comprehensive dental medicine and expand the scope of ongoing large-scale research projects.

Deep Learning Ensemble Method for Classifying Glaucoma Stages Using Fundus Photographs and Convolutional Neural Networks.

Purpose: This study developed and evaluated a deep learning ensemble method to automatically grade the stages of glaucoma depending on its severity.Materials and Methods: After cross-validation of three glaucoma specialists, the final dataset comprised of 3,460 fundus photographs taken from 2,204 patients were divided into three classes: unaffected controls, early-stage glaucoma, and late-stage glaucoma. The mean deviation value of standard automated perimetry was used to classify the glaucoma cases. We modeled 56 convolutional neural networks (CNN) with different characteristics and developed an ensemble system to derive the best performance by combining several modeling results.Results: The proposed method with an accuracy of 88.1% and an average area under the receiver operating characteristic of 0.975 demonstrates significantly better performance to classify glaucoma stages compared to the best single CNN model that has an accuracy of 85.2% and an average area under the receiver operating characteristic of 0.950. The false negative is the least adjacent misprediction, and it is less in the proposed method than in the best single CNN model.Conclusions: The method of averaging multiple CNN models can better classify glaucoma stages by using fundus photographs than a single CNN model. The ensemble method would be useful as a clinical decision support system in glaucoma screening for primary care because it provides high and stable performance with a relatively small amount of data.

Early Prediction of the Need for Orthognathic Surgery in Patients With Repaired Unilateral Cleft Lip and Palate Using Machine Learning and Longitudinal Lateral Cephalometric Analysis Data.

ABSTRACT: The purpose of this study was to determine the cephalometric predictors of the future need for orthognathic surgery in patients with repaired unilateral cleft lip and palate (UCLP) using machine learning. This study included 56 Korean patients with UCLP, who were treated by a single surgeon and a single orthodontist with the same treatment protocol. Lateral cephalograms were obtained before the commencement of orthodontic/orthopedic treatment (T0; mean age, 6.3 years) and at at least of 15 years of age (T1; mean age, 16.7 years). 38 cephalometric variables were measured. At T1 stage, 3 cephalometric criteria (ANB ≤ -3°; Wits appraisal ≤ -5 mm; Harvold unit difference ≥34 mm for surgery group) were used to classify the subjects into the surgery group (n = 10, 17.9%) and non-surgery group (n = 46, 82.1%). Independent t-test was used for statistical analyses. The Boruta method and XGBoost algorithm were used to determine the cephalometric variables for the prediction model. At T0 stage, 2 variables exhibited a significant intergroup difference (ANB and facial convexity angle [FCA], all P < 0.05). However, 18 cephalometric variables at the T1 stage and 14 variables in the amount of change (ΔT1-T0) exhibited significant intergroup differences (all, more significant than P < 0.05). At T0 stage, the ANB, PP-FH, combination factor, and FCA were selected as predictive parameters with a cross-validation accuracy of 87.4%. It was possible to predict the future need for surgery to correct sagittal skeletal discrepancy in UCLP patients at the age of 6 years.

An Organoid Biobank of Neuroendocrine Neoplasms Enables Genotype-Phenotype Mapping.

Gastroenteropancreatic (GEP) neuroendocrine neoplasm (NEN) that consists of neuroendocrine tumor and neuroendocrine carcinoma (NEC) is a lethal but under-investigated disease owing to its rarity. To fill the scarcity of clinically relevant models of GEP-NEN, we here established 25 lines of NEN organoids and performed their comprehensive molecular characterization. GEP-NEN organoids recapitulated pathohistological and functional phenotypes of the original tumors. Whole-genome sequencing revealed frequent genetic alterations in TP53 and RB1 in GEP-NECs, and characteristic chromosome-wide loss of heterozygosity in GEP-NENs. Transcriptome analysis identified molecular subtypes that are distinguished by the expression of distinct transcription factors. GEP-NEN organoids gained independence from the stem cell niche irrespective of genetic mutations. Compound knockout of TP53 and RB1, together with overexpression of key transcription factors, conferred on the normal colonic epithelium phenotypes that are compatible with GEP-NEN biology. Altogether, our study not only provides genetic understanding of GEP-NEN, but also connects its genetics and biological phenotypes.

Association of HSV-1 and Reduced Oral Bacteriota Diversity with Chemotherapy-Induced Oral Mucositis in Patients Undergoing Autologous Hematopoietic Stem Cell Transplantation.

Oral mucositis (OM) is a common complication of chemotherapy and remains a significant unmet need. The aim of this study was to investigate the role of oral bacteriota and HSV-1 in OM. Forty-six patients admitted for autologous hematopoietic stem cell transplantation were longitudinally evaluated for OM, Candida, HSV-1, and leukocyte count, and buccal mucosal bacterial samples were obtained during their admission period. The bacterial communities collected at the baseline and post-chemotherapy, chosen from the time with the highest severity, were analyzed by sequencing the 16S rRNA gene. Twenty (43.5%) patients developed OM, the severity of which ranged from 1 to 5 according to the Oral Mucositis Assessment Scale (OMAS). Chemotherapy significantly increased the prevalence of HSV-1 detection but not that of Candida. The bacterial communities of patients after conditioning chemotherapy were characterized by aberrant enrichment of minor species and decreased evenness and Shannon diversity. After adjustment for age, gender, and neutropenia, the presence of HSV-1 was associated with the incidence of OM (odds ratio = 3.668, p = 0.004), while the decrease in Shannon diversity was associated with the severity of OM (ß = 0.533 ± 0.220, p = 0.015). The control of HSV-1 and restoration of oral bacterial diversity may be a novel option to treat or prevent OM.

Analysis of brain connectivity during nitrous oxide sedation using graph theory.

Nitrous oxide, the least potent inhalation anesthetic, is widely used for conscious sedation. Recently, it has been reported that the occurrence of anesthetic-induced loss of consciousness decreases the interconnection between brain regions, resulting in brain network changes. However, few studies have investigated these changes in conscious sedation using nitrous oxide. Therefore, the present study aimed to use graph theory to analyze changes in brain networks during nitrous oxide sedation. Participants were 20 healthy volunteers (10 men and 10 women, 20-40 years old) with no history of systemic disease. We acquired electroencephalogram (EEG) recordings of 32 channels during baseline, nitrous oxide inhalation sedation, and recovery. EEG epochs from the baseline and the sedation state (50% nitrous oxide) were extracted and analyzed with the network connection parameters of graph theory. Analysis of 1/f dynamics, revealed a steeper slope while in the sedation state than during the baseline. Network connectivity parameters showed significant differences between the baseline and sedation state, in delta, alpha1, alpha2, and beta2 frequency bands. The most pronounced differences in functional distance during nitrous oxide sedation were observed in the alpha1 and alpha2 frequency bands. Change in 1/f dynamics indicates that changes in brain network systems occur during nitrous oxide administration. Changes in network parameters imply that nitrous oxide interferes with the efficiency of information integration in the frequency bands important for cognitive processes and attention tasks. Alteration of brain network during nitrous oxide administration may be associated to the sedative mechanism of nitrous oxide.

Somatic mutation landscape reveals differential variability of cell-of-origin for primary liver cancer.

Primary liver tissue cancer types are renowned to display a consistent increase in global disease burden and mortality, thus needing more effective diagnostics and treatments. Yet, integrative research efforts to identify cell-of-origin for these cancers by utilizing human specimen data were poorly established. To this end, we analyzed previously published whole-genome sequencing data for 384 tumor and progenitor tissues along with 423 publicly available normal tissue epigenomic features and single cell RNA-seq data from human livers to assess correlation patterns and extended this information to conduct in-silico prediction of the cell-of-origin for primary liver cancer subtypes. Despite mixed histological features, the cell-of-origin for mixed hepatocellular carcinoma/intrahepatic cholangiocarcinoma subtype was predominantly predicted to be hepatocytic origin. Individual sample-level predictions also revealed hepatocytes as one of the major predicted cell-of-origin for intrahepatic cholangiocarcinoma, thus implying trans-differentiation process during cancer progression. Additional analyses on the whole genome sequencing data of hepatic progenitor cells suggest these cells may not be a direct cell-of-origin for liver cancers. These results provide novel insights on the nature and potential contributors of cell-of-origins for primary liver cancers.

Altered expression of fucosylation pathway genes is associated with poor prognosis and tumor metastasis in non­small cell lung cancer.

Fucosylation is a post­translational modification that attaches fucose residues to protein­ or lipid­bound oligosaccharides. Certain fucosylation pathway genes are aberrantly expressed in several types of cancer, including non­small cell lung cancer (NSCLC), and this aberrant expression is associated with poor prognosis in patients with cancer. However, the molecular mechanism by which these fucosylation pathway genes promote tumor progression has not been well­characterized. The present study analyzed public microarray data obtained from NSCLC samples. Multivariate analysis revealed that altered expression of fucosylation pathway genes, including fucosyltransferase 1 (FUT1), FUT2, FUT3, FUT6, FUT8 and GDP­L­fucose synthase (TSTA3), correlated with poor survival in patients with NSCLC. Inhibition of FUTs by 2F­peracetyl­fucose (2F­PAF) suppressed transforming growth factor ß (TGFß)­mediated Smad3 phosphorylation and nuclear translocation in NSCLC cells. In addition, wound­healing and Transwell migration assays demonstrated that 2F­PAF inhibited TGFß­induced NSCLC cell migration and invasion. Furthermore, in vivo bioluminescence imaging analysis revealed that 2F­PAF attenuated the metastatic capacity of NSCLC cells. These results may help characterize the oncogenic role of fucosylation in NSCLC biology and highlight its potential for developing cancer therapeutics.

Transcriptome and methylome analysis of periodontitis and peri-implantitis with tobacco use.

Peri-implantitis is similar to periodontitis in both symptoms and treatment; however, their level of similarity remains controversial. Here, we compared multiple cases of periodontitis and peri-implantitis through transcriptome and methylome profiling, and analyzed the effects of smoking as a typical risk factor. Human gingival tissues were obtained from 20 patients with periodontitis or peri-implantitis via periodontal surgical procedures. Total RNA and genomic DNA were isolated, and transcriptome and methylome datasets were generated. Comprehensive analysis of differential gene expression, DNA methylation, and protein-protein interactions indicated that periodontitis and peri-implantitis share biological similarities; however, hierarchical clustering between the two disease groups revealed distinct molecular characteristics. These differences might be related to structural differences in natural tooth-bone and implant-bone. Additionally, smoking differentially affected periodontitis and peri-implantitis in terms of host-defense mechanism impairment. Within the limitations of this study, the results provide evidence that peri-implantitis is distinct from periodontitis and that smoking potentially affects disease progression. Our study provides a foundation for the rational design of a large-scale study in the future for a more comprehensive analysis that includes microbiome and clinical data.

Multi-channel PINN: investigating scalable and transferable neural networks for drug discovery.

Analysis of compound-protein interactions (CPIs) has become a crucial prerequisite for drug discovery and drug repositioning. In vitro experiments are commonly used in identifying CPIs, but it is not feasible to discover the molecular and proteomic space only through experimental approaches. Machine learning's advances in predicting CPIs have made significant contributions to drug discovery. Deep neural networks (DNNs), which have recently been applied to predict CPIs, performed better than other shallow classifiers. However, such techniques commonly require a considerable volume of dense data for each training target. Although the number of publicly available CPI data has grown rapidly, public data is still sparse and has a large number of measurement errors. In this paper, we propose a novel method, Multi-channel PINN, to fully utilize sparse data in terms of representation learning. With representation learning, Multi-channel PINN can utilize three approaches of DNNs which are a classifier, a feature extractor, and an end-to-end learner. Multi-channel PINN can be fed with both low and high levels of representations and incorporates each of them by utilizing all approaches within a single model. To fully utilize sparse public data, we additionally explore the potential of transferring representations from training tasks to test tasks. As a proof of concept, Multi-channel PINN was evaluated on fifteen combinations of feature pairs to investigate how they affect the performance in terms of highest performance, initial performance, and convergence speed. The experimental results obtained indicate that the multi-channel models using protein features performed better than single-channel models or multi-channel models using compound features. Therefore, Multi-channel PINN can be advantageous when used with appropriate representations. Additionally, we pretrained models on a training task then finetuned them on a test task to figure out whether Multi-channel PINN can capture general representations for compounds and proteins. We found that there were significant differences in performance between pretrained models and non-pretrained models.