Interpreting drug synergy in breast cancer with deep learning using target-protein inhibition profiles

BACKGROUND

Breast cancer is the most common malignancy among women worldwide. Despite advances in treating breast cancer over the past decades, drug resistance and adverse effects remain challenging. Recent therapeutic progress has shifted toward using drug combinations for better treatment efficiency. However, with a growing number of potential small-molecule cancer inhibitors, in silico strategies to predict pharmacological synergy before experimental trials are required to compensate for time and cost restrictions. Many deep learning models have been previously proposed to predict the synergistic effects of drug combinations with high performance. However, these models heavily relied on a large number of drug chemical structural fingerprints as their main features, which made model interpretation a challenge.

RESULTS

This study developed a deep neural network model that predicts synergy between small-molecule pairs based on their inhibitory activities against 13 selected key proteins. The synergy prediction model achieved a Pearson correlation coefficient between model predictions and experimental data of 0.63 across five breast cancer cell lines. BT-549 and MCF-7 achieved the highest correlation of 0.67 when considering individual cell lines. Despite achieving a moderate correlation compared to previous deep learning models, our model offers a distinctive advantage in terms of interpretability. Using the inhibitory activities against key protein targets as the main features allowed a straightforward interpretation of the model since the individual features had direct biological meaning. By tracing the synergistic interactions of compounds through their target proteins, we gained insights into the patterns our model recognized as indicative of synergistic effects.

CONCLUSIONS

The framework employed in the present study lays the groundwork for future advancements, especially in model interpretation. By combining deep learning techniques and target-specific models, this study shed light on potential patterns of target-protein inhibition profiles that could be exploited in breast cancer treatment.

Boolean modeling of breast cancer signaling pathways uncovers mechanisms of drug synergy

Breast cancer is one of the most common types of cancer in females. While drug combinations have shown potential in breast cancer treatments, identifying new effective drug pairs is challenging due to the vast number of possible combinations among available compounds. Efforts have been made to accelerate the process with in silico predictions. Here, we developed a Boolean model of signaling pathways in breast cancer. The model was tailored to represent five breast cancer cell lines by integrating information about cell-line specific mutations, gene expression, and drug treatments. The models reproduced cell-line specific protein activities and drug-response behaviors in agreement with experimental data. Next, we proposed a calculation of protein synergy scores (PSSs), determining the effect of drug combinations on individual proteins' activities. The PSSs of selected proteins were used to investigate the synergistic effects of 150 drug combinations across five cancer cell lines. The comparison of the highest single agent (HSA) synergy scores between experiments and model predictions from the MDA-MB-231 cell line achieved the highest Pearson's correlation coefficient of 0.58 with a great balance among the classification metrics (AUC = 0.74, sensitivity = 0.63, and specificity = 0.64). Finally, we clustered drug pairs into groups based on the selected PSSs to gain further insights into the mechanisms underlying the observed synergistic effects of drug pairs. Clustering analysis allowed us to identify distinct patterns in the protein activities that correspond to five different modes of synergy: 1) synergistic activation of FADD and BID (extrinsic apoptosis pathway), 2) synergistic inhibition of BCL2 (intrinsic apoptosis pathway), 3) synergistic inhibition of MTORC1, 4) synergistic inhibition of ESR1, and 5) synergistic inhibition of CYCLIN D. Our approach offers a mechanistic understanding of the efficacy of drug combinations and provides direction for selecting potential drug pairs worthy of further laboratory investigation.

mSRFR: a machine learning model using microalgal signature features for ncRNA classification

This work presents mSRFR (microalgae SMOTE Random Forest Relief model), a classification tool for noncoding RNAs (ncRNAs) in microalgae, including green algae, diatoms, golden algae, and cyanobacteria. First, the SMOTE technique was applied to address the challenge of imbalanced data due to the different numbers of microalgae ncRNAs from different species in the EBI RNA-central database. Then the top 20 significant features from a total of 106 features, including sequence-based, secondary structure, base-pair, and triplet sequence-structure features, were selected using the Relief feature selection method. Next, ten-fold cross-validation was applied to choose a classifier algorithm with the highest performance among Support Vector Machine, Random Forest, Decision Tree, Naïve Bayes, K-nearest Neighbor, and Neural Network, based on the receiver operating characteristic (ROC) area. The results showed that the Random Forest classifier achieved the highest ROC area of 0.992. Then, the Random Forest algorithm was selected and compared with other tools, including RNAcon, CPC, CPC2, CNCI, and CPPred. Our model achieved a high accuracy of about 97% and a low false-positive rate of about 2% in predicting the test dataset of microalgae. Furthermore, the top features from Relief revealed that the %GA dinucleotide is a signature feature of microalgal ncRNAs when compared to Escherichia coli, Saccharomyces cerevisiae, Arabidopsis thaliana, and Homo sapiens.

Immunodominant and Neutralizing Linear B-Cell Epitopes Spanning the Spike and Membrane Proteins of Porcine Epidemic Diarrhea Virus

Porcine epidemic diarrhea virus (PEDV) is the causative agent of PED, an enteric disease that causes high mortality rates in piglets. PEDV is an alphacoronavirus that has high genetic diversity. Insights into neutralizing B-cell epitopes of all genetically diverse PEDV strains are of importance, particularly for designing a vaccine that can provide broad protection against PEDV. In this work, we aimed to explore the landscape of linear B-cell epitopes on the spike (S) and membrane (M) proteins of global PEDV strains. All amino acid sequences of the PEDV S and M proteins were retrieved from the NCBI database and grouped. Immunoinformatics-based methods were next developed and used to identify putative linear B-cell epitopes from 14 and 5 consensus sequences generated from distinct groups of the S and M proteins, respectively. ELISA testing predicted peptides with PEDV-positive sera revealed nine novel immunodominant epitopes on the S protein. Importantly, seven of these novel immunodominant epitopes and other subdominant epitopes were demonstrated to be neutralizing epitopes by neutralization–inhibition assay. Our findings unveil important roles of the PEDV S2 subunit in both immune stimulation and virus neutralization. Additionally, our study shows the first time that the M protein is also the target of PEDV neutralization with seven neutralizing epitopes identified. Conservancy profiles of the epitopes are also provided. In this study, we offer immunoinformatics-based methods for linear B-cell epitope identification and a more complete profile of linear B-cell epitopes across the PEDV S and M proteins, which may contribute to the development of a greater next-generation PEDV vaccine as well as peptide-based immunoassays.

Construction of Light-Responsive Gene Regulatory Network for Growth, Development and Secondary Metabolite Production in Cordyceps militaris

Cordyceps militaris is an edible fungus that produces many beneficial compounds, including cordycepin and carotenoid. In many fungi, growth, development and secondary metabolite production are controlled by crosstalk between light-signaling pathways and other regulatory cascades. However, little is known about the gene regulation upon light exposure in C. militaris. This study aims to construct a gene regulatory network (GRN) that responds to light in C. militaris. First, a genome-scale GRN was built based on transcription factor (TF)-target gene interactions predicted from the Regulatory Sequence Analysis Tools (RSAT). Then, a light-responsive GRN was extracted by integrating the transcriptomic data onto the genome-scale GRN. The light-responsive network contains 2689 genes and 6837 interactions. From the network, five TFs, Snf21 (CCM_04586), an AT-hook DNA-binding motif TF (CCM_08536), a homeobox TF (CCM_07504), a forkhead box protein L2 (CCM_02646) and a heat shock factor Hsf1 (CCM_05142), were identified as key regulators that co-regulate a large group of growth and developmental genes. The identified regulatory network and expression profiles from our analysis suggested how light may induce the growth and development of C. militaris into a sexual cycle. The light-mediated regulation also couples fungal development with cordycepin and carotenoid production. This study leads to an enhanced understanding of the light-responsive regulation of growth, development and secondary metabolite production in the fungi.

Immunodominant linear B cell epitopes in the spike and membrane proteins of SARS-CoV-2 identified by immunoinformatics prediction and immunoassay

SARS-CoV-2 continues to infect an ever-expanding number of people, resulting in an increase in the number of deaths globally. With the emergence of new variants, there is a corresponding decrease in the currently available vaccine efficacy, highlighting the need for greater insights into the viral epitope profile for both vaccine design and assessment. In this study, three immunodominant linear B cell epitopes in the SARS-CoV-2 spike receptor-binding domain (RBD) were identified by immunoinformatics prediction, and confirmed by ELISA with sera from Macaca fascicularis vaccinated with a SARS-CoV-2 RBD subunit vaccine. Further immunoinformatics analyses of these three epitopes gave rise to a method of linear B cell epitope prediction and selection. B cell epitopes in the spike (S), membrane (M), and envelope (E) proteins were subsequently predicted and confirmed using convalescent sera from COVID-19 infected patients. Immunodominant epitopes were identified in three regions of the S2 domain, one region at the S1/S2 cleavage site and one region at the C-terminus of the M protein. Epitope mapping revealed that most of the amino acid changes found in variants of concern are located within B cell epitopes in the NTD, RBD, and S1/S2 cleavage site. This work provides insights into B cell epitopes of SARS-CoV-2 as well as immunoinformatics methods for B cell epitope prediction, which will improve and enhance SARS-CoV-2 vaccine development against emergent variants.

Natural ACE inhibitory peptides discovery from Spirulina (Arthrospira platensis) strain C1

Bioactive peptides from natural sources are utilized as food supplements for disease prevention and are increasingly becoming targets for drug discovery due to their specificity, efficacy and the absence of undesirable side effects, among others. Hence, the 'SpirPep' platform was developed to facilitate the in silico-based bioactive peptide discovery of these highly sought-after biomolecules from Spirulina(Arthrospira platensis) and to select the protease (thermolysin) used for in vitro digestion. Analysis of the predicted and experimentally-derived peptides suggested that they were mainly involved in ACE inhibition; thus, an ACEi assay was used to study the ACE inhibitory activity of five candidate peptides (SpirPep1-5), chosen from common peptides with multifunctional bioactivity and 100% bioactive peptide coverage, originating from phycobiliproteins. Results showed that SpirPep1 inhibited the activity of ACE with IC₅₀ of 1.748 mM and was non-toxic to fibroblasts of African green monkey kidney and human dermal skin. The molecular docking and MD simulation analysis revealed SpirPep1 had significantly lower binding scores than others and showed greater specificity to ACE. The non-bonded interaction energy of SpirPep1 and ACE was -883 kJ/mol. The SpirPep1 indirectly bound to ACE via the ACE substrate binding sites residues (D121, E123, S516, and S517) found in natural ACE inhibitory peptides (angiotensin II and bradykinin potentiating peptides). In addition, two unreported substrate binding sites including R124 and S219 were found. These results indicate that 'SpirPep' platform could increase the success rate for natural bioactive peptide discovery.

SpirPep: an in silico digestion-based platform to assist bioactive peptides discovery from a genome-wide database

Background

Bioactive peptides, including biological sources-derived peptides with different biological activities, are protein fragments that influence the functions or conditions of organisms, in particular humans and animals. Conventional methods of identifying bioactive peptides are time-consuming and costly. To quicken the processes, several bioinformatics tools are recently used to facilitate screening of the potential peptides prior their activity assessment in vitro and/or in vivo. In this study, we developed an efficient computational method, SpirPep, which offers many advantages over the currently available tools.

Results

The SpirPep web application tool is a one-stop analysis and visualization facility to assist bioactive peptide discovery. The tool is equipped with 15 customized enzymes and 1–3 miscleavage options, which allows in silico digestion of protein sequences encoded by protein-coding genes from single, multiple, or genome-wide scaling, and then directly classifies the peptides by bioactivity using an in-house database that contains bioactive peptides collected from 13 public databases. With this tool, the resulting peptides are categorized by each selected enzyme, and shown in a tabular format where the peptide sequences can be tracked back to their original proteins. The developed tool and webpages are coded in PHP and HTML with CSS/JavaScript. Moreover, the tool allows protein-peptide alignment visualization by Generic Genome Browser (GBrowse) to display the region and details of the proteins and peptides within each parameter, while considering digestion design for the desirable bioactivity. SpirPep is efficient; it takes less than 20 min to digest 3000 proteins (751,860 amino acids) with 15 enzymes and three miscleavages for each enzyme, and only a few seconds for single enzyme digestion. Obviously, the tool identified more bioactive peptides than that of the benchmarked tool; an example of validated pentapeptide (FLPIL) from LC-MS/MS was demonstrated. The web and database server are available at http://spirpepapp.sbi.kmutt.ac.th.

Conclusion

SpirPep, a web-based bioactive peptide discovery application, is an in silico-based tool with an overview of the results. The platform is a one-stop analysis and visualization facility; and offers advantages over the currently available tools. This tool may be useful for further bioactivity analysis and the quantitative discovery of desirable peptides.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2143-0) contains supplementary material, which is available to authorized users.

Molecular dynamics exploration of poration and leaking caused by Kalata B1 in HIV-infected cell membrane compared to host and HIV membranes

The membrane disruption activities of kalata B1 (kB1) were investigated using molecular dynamics simulations with membrane models. The models were constructed to mimic the lipid microdomain formation in membranes of HIV particle, HIV-infected cell, and host cell. The differences in the lipid ratios of these membranes caused the formation of liquid ordered (lo) domains of different sizes, which affected the binding and activity of kB1. Stronger kB1 disruptive activity was observed for the membrane with small sized lo domain. Our results show that kB1 causes membrane leaking without bilayer penetration. The membrane poration mechanism involved in the disorganization of the lo domain and in cholesterol inter-leaflet translocation is described. This study enhances our understanding of the membrane activity of kB1, which may be useful for designing novel and potentially therapeutic peptides based on the kB1 framework.

Computational Design of Hypothetical New Peptides Based on a Cyclotide Scaffold as HIV gp120 Inhibitor

Cyclotides are a family of triple disulfide cyclic peptides with exceptional resistance to thermal/chemical denaturation and enzymatic degradation. Several cyclotides have been shown to possess anti-HIV activity, including kalata B1 (KB1). However, the use of cyclotides as anti-HIV therapies remains limited due to the high toxicity in normal cells. Therefore, grafting anti-HIV epitopes onto a cyclotide might be a promising approach for reducing toxicity and simultaneously improving anti-HIV activity. Viral envelope glycoprotein gp120 is required for entry of HIV into CD4+ T cells. However, due to a high degree of variability and physical shielding, the design of drugs targeting gp120 remains challenging. We created a computational protocol in which molecular modeling techniques were combined with a genetic algorithm (GA) to automate the design of new cyclotides with improved binding to HIV gp120. We found that the group of modified cyclotides has better binding scores (23.1%) compared to the KB1. By using molecular dynamic (MD) simulation as a post filter for the final candidates, we identified two novel cyclotides, GA763 and GA190, which exhibited better interaction energies (36.6% and 22.8%, respectively) when binding to gp120 compared to KB1. This computational design represents an alternative tool for modifying peptides, including cyclotides and other stable peptides, as therapeutic agents before the synthesis process.

Dynamic scenario of membrane binding process of kalata b1

Kalata B1 (kB1), a cyclotide that has been used in medical applications, displays cytotoxicity related to membrane binding and oligomerization. Our molecular dynamics simulation results demonstrate that Trp19 in loop 5 of both monomeric and tetrameric kB1 is a key residue for initial anchoring in the membrane binding process. This residue also facilitates the formation of kB1 tetramers. Additionally, we elucidate that kB1 preferentially binds to the membrane interfacial zone and is unable to penetrate into the membrane. In particular, significant roles of amino acid residues in loop 5 and loop 6 on the localization of kB1 to this membrane-water interface zone are found. This study reveals the roles of amino acid residues in the bioactivity of kB1, which is information that can be useful for designing new therapeutic cyclotides with less toxicity.

Identification of non-coding RNAs with a new composite feature in the Hybrid Random Forest Ensemble algorithm

To identify non-coding RNA (ncRNA) signals within genomic regions, a classification tool was developed based on a hybrid random forest (RF) with a logistic regression model to efficiently discriminate short ncRNA sequences as well as long complex ncRNA sequences. This RF-based classifier was trained on a well-balanced dataset with a discriminative set of features and achieved an accuracy, sensitivity and specificity of 92.11%, 90.7% and 93.5%, respectively. The selected feature set includes a new proposed feature, SCORE. This feature is generated based on a logistic regression function that combines five significant features-structure, sequence, modularity, structural robustness and coding potential-to enable improved characterization of long ncRNA (lncRNA) elements. The use of SCORE improved the performance of the RF-based classifier in the identification of Rfam lncRNA families. A genome-wide ncRNA classification framework was applied to a wide variety of organisms, with an emphasis on those of economic, social, public health, environmental and agricultural significance, such as various bacteria genomes, the Arthrospira (Spirulina) genome, and rice and human genomic regions. Our framework was able to identify known ncRNAs with sensitivities of greater than 90% and 77.7% for prokaryotic and eukaryotic sequences, respectively. Our classifier is available at http://ncrna-pred.com/HLRF.htm.

Heterogeneous ensemble approach with discriminative features and modified-SMOTEbagging for pre-miRNA classification

An ensemble classifier approach for microRNA precursor (pre-miRNA) classification was proposed based upon combining a set of heterogeneous algorithms including support vector machine (SVM), k-nearest neighbors (kNN) and random forest (RF), then aggregating their prediction through a voting system. Additionally, the proposed algorithm, the classification performance was also improved using discriminative features, self-containment and its derivatives, which have shown unique structural robustness characteristics of pre-miRNAs. These are applicable across different species. By applying preprocessing methods--both a correlation-based feature selection (CFS) with genetic algorithm (GA) search method and a modified-Synthetic Minority Oversampling Technique (SMOTE) bagging rebalancing method--improvement in the performance of this ensemble was observed. The overall prediction accuracies obtained via 10 runs of 5-fold cross validation (CV) was 96.54%, with sensitivity of 94.8% and specificity of 98.3%-this is better in trade-off sensitivity and specificity values than those of other state-of-the-art methods. The ensemble model was applied to animal, plant and virus pre-miRNA and achieved high accuracy, >93%. Exploiting the discriminative set of selected features also suggests that pre-miRNAs possess high intrinsic structural robustness as compared with other stem loops. Our heterogeneous ensemble method gave a relatively more reliable prediction than those using single classifiers. Our program is available at http://ncrna-pred.com/premiRNA.html.

Draft genome sequence of Arthrospira platensis C1 (PCC9438)

Arthrospira platensis is a cyanobacterium that is extensively cultivated outdoors on a large commercial scale for consumption as a food for humans and animals. It can be grown in monoculture under highly alkaline conditions, making it attractive for industrial production. Here we describe the complete genome sequence of A. platensis C1 strain and its annotation. The A. platensis C1 genome contains 6,089,210 bp including 6,108 protein-coding genes and 45 RNA genes, and no plasmids. The genome information has been used for further comparative analysis, particularly of metabolic pathways, photosynthetic efficiency and barriers to gene transfer.

Functional expression of Spirulina-Delta6 desaturase gene in yeast, Saccharomyces cerevisiae

Spirulina-acyl-lipid desaturases are membrane-bound enzymes found in thylakoid and plasma membranes. These enzymes carry out the fatty acid desaturation process of Spirulina to yield gamma-linolenic acid (GLA) as the final desaturation product. In this study, Spirulina-Delta(6) desaturase encoded by the desD gene was heterologously expressed and characterized in Saccharomyces cerevisiae. We then conducted site-directed mutagenesis of the histidine residues in the three histidine boxes to determine the role of these amino acid residues in the enzyme function. Our results showed that while four mutants showed complete loss of Delta(6)-desaturase activity and two mutants showed only trace of the activity, the enzyme activity could be partially restored by chemical rescue using exogenously provided imidazole. This study reveals that the histidine residues (which have imidazole as their functional group) in the conserved clusters play a critical role in Delta(6)-desaturase activity, possibly by providing a di-iron catalytic center. In our previous study, this enzyme was expressed in Escherichia coli. The results reveal that the enzyme can function only in the presence of an exogenous cofactor, ferredoxin, provided in vitro. This evidence suggests that baker's yeast has a cofactor that can complement ferredoxin, thought to act as an electron donor for the Delta(6) desaturation in cyanobacteria, including Spirulina. The electron donor of the Spirulina-Delta(6) desaturation in yeast is more likely to be cytochrome b(5), which is absent in E. coli. This means that the enzyme expressed in S. cerevisiae can catalyze the biosynthesis of the product, GLA, in vivo.

Higher plant-like fluorescence induction and thermoluminescence characteristics in cyanobacterium, Spirulina mutant defective in PQH2 oxidation by cytb6/f complex

Characterization of the photosynthetic electron transport in a mutant of Spirulina platensis, generated by chemical mutagenesis, demonstrated that the electron transfer from the plastoquinone (PQ) to cytochrome b6/f was slowed. Thermoluminescence (TL) measurements suggested the presence of reversed energy flow via PQ, which resulted in an emergence of the plant-like after-glow TL band at 45 degrees C that could be enhanced by the transthylakoidal pH gradient and could be eliminated by an uncoupler, FCCP. The localization of the changes in the electron transport of the mutant cells measured by various methods revealed that the re-oxidation of the PQ pool is hampered in the mutant compared to the wild-type cells. The reduction in energy migration was localized between PQ and PS I reaction centers.

Purification and characterization of the IM-2-binding protein from Streptomyces sp. strain FRI-5

IM-2 [(2R,3R,1'R)-2-(1'-hydroxybutyl)-3-(hydroxymethyl)butanolide] of Streptomyces sp. strain FRI-5 is one of the butyrolactone autoregulators of Streptomyces species and triggers production of blue pigment as well as the nucleoside antibiotics showdomycin and minimycin. A tritium-labeled IM-2 analogue, 2,3-trans-2(1'-beta-hydroxy-[4',5'-3H]pentyl)-3-(hydroxymethyl)butano lide ([3H]IM-2-C5; 40 Ci/mmol), was synthesized for a competitive binding assay, and an IM-2-specific binding protein was found to be present in the crude cell extract of Streptomyces sp. strain FRI-5. During cultivation for 24 h, the specific IM-2-binding activity increased rapidly, reached a plateau at 10 to 14 h, and declined sharply thereafter, showing only 6% activity after 24 h of cultivation. A Scatchard plot of the binding data demonstrated that the dissociation constant (Kd) for [3H]IM-2-C5 was 1.3 nM, while the Kd for a 3H-labeled virginiae butanolide (VB) analogue, 2-(1'-alpha-hydroxy-[6',7'-3H]heptyl)-3-(hydroxymethyl)butanolide ([3H]VB-C7), another butyrolactone autoregulator possessing the opposite configuration at C-1' was 35 nM. Furthermore, at a 15-fold molar excess, the effectiveness of several autoregulators as nonlabeled competitive ligands against [3H]IM-2-C5 was IM-2 type > VB-C type >> A-factor type, indicating that the binding protein in Streptomyces sp. strain FRI-5 is highly specific toward IM-2. Ultracentrifugation showed that the IM-2-binding protein is present almost exclusively in the 100,000 x g supernatant fraction, indicating that the binding protein is a cytoplasmic soluble protein. The binding protein was purified by ammonium sulfate precipitation, DEAE-Sephacel chromatography, Sephacryl S-100 HR gel filtration, DEAE-5PW high-performance liquid chromatography (HPLC), and phenyl-5PW HPLC. The apparent Mr of the native IM-2-binding protein as determined by molecular sieve HPLC was about 60,000 in the presence of 0.5, 0.3, or 0.1 M KCl, while by sodium dodecyl sulfate-polyacrylamide gel electrophoresis it was about 27,000, suggesting that the native binding protein is present in the form of a dimer.