Using Machine Learning and miRNA for the Diagnosis of Esophageal Cancer.

BACKGROUND: Esophageal cancer (EC) remains a global health challenge, often diagnosed at advanced stages, leading to high mortality rates. Current diagnostic tools for EC are limited in their efficacy. This study aims to harness the potential of microRNAs (miRNAs) as novel, noninvasive diagnostic biomarkers for EC. Our objective was to determine the diagnostic accuracy of miRNAs, particularly in distinguishing miRNAs associated with EC from control miRNAs. METHODS: We applied machine learning (ML) techniques in WEKA (Waikato Environment for Knowledge Analysis) and TensorFlow Keras to a dataset of miRNA sequences and gene targets, assessing the predictive power of several classifiers: naïve Bayes, multilayer perceptron, Hoeffding tree, random forest, and random tree. The data were further subjected to InfoGain feature selection to identify the most informative miRNA sequence and gene target descriptors. The ML models' abilities to distinguish between miRNA implicated in EC and control group miRNA was then tested. RESULTS: Of the tested WEKA classifiers, the top 3 performing ones were random forest, Hoeffding tree, and naïve Bayes. The TensorFlow Keras neural network model was subsequently trained and tested, the model's predictive power was further validated using an independent dataset. The TensorFlow Keras gave an accuracy 0.91. The WEKA best algorithm (naïve Bayes) model yielded an accuracy of 0.94. CONCLUSIONS: The results demonstrate the potential of ML-based miRNA classifiers in diagnosing EC. However, further studies are necessary to validate these findings and explore the full clinical potential of this approach.

Colorectal Cancer Detection via Metabolites and Machine Learning.

Today, colorectal cancer (CRC) diagnosis is performed using colonoscopy, which is the current, most effective screening method. However, colonoscopy poses risks of harm to the patient and is an invasive process. Recent research has proven metabolomics as a potential, non-invasive detection method, which can use identified biomarkers to detect potential cancer in a patient's body. The aim of this study is to develop a machine-learning (ML) model based on chemical descriptors that will recognize CRC-associated metabolites. We selected a set of metabolites found as the biomarkers of CRC, confirmed that they participate in cancer-related pathways, and used them for training a machine-learning model for the diagnostics of CRC. Using a set of selective metabolites and random compounds, we developed a range of ML models. The best performing ML model trained on Stage 0-2 CRC metabolite data predicted a metabolite class with 89.55% accuracy. The best performing ML model trained on Stage 3-4 CRC metabolite data predicted a metabolite class with 95.21% accuracy. Lastly, the best-performing ML model trained on Stage 0-4 CRC metabolite data predicted a metabolite class with 93.04% accuracy. These models were then tested on independent datasets, including random and unrelated-disease metabolites. In addition, six pathways related to these CRC metabolites were also distinguished: aminoacyl-tRNA biosynthesis; glyoxylate and dicarboxylate metabolism; glycine, serine, and threonine metabolism; phenylalanine, tyrosine, and tryptophan biosynthesis; arginine biosynthesis; and alanine, aspartate, and glutamate metabolism. Thus, in this research study, we created machine-learning models based on metabolite-related descriptors that may be helpful in developing a non-invasive diagnosis method for CRC.

Laryngeal cancer diagnosis via miRNA-based decision tree model.

PURPOSE: Laryngeal cancer (LC) is the most common head and neck cancer, which often goes undiagnosed due to the inaccessible nature of current diagnosis methods in some parts of the world. Many recent studies have shown that microRNAs (miRNAs) are crucial biomarkers for a variety of cancers. METHODS: In this study, we create a decision tree model for the diagnosis of laryngeal cancer using a created series of miRNA attributes, such as sequence-based characteristics, predicted miRNA target genes, and gene pathways. This series of attributes is extracted from both differentially expressed blood-based miRNAs in laryngeal cancer and random, non-associated with cancer miRNAs. RESULTS: Several machine-learning (ML) algorithms were tested in the ML model, and the Hoeffding Tree classifier yields the highest accuracy (86.8%) in miRNAs-based recognition of laryngeal cancer. Furthermore, our model is validated with the independent laryngeal cancer datasets and can accurately diagnose laryngeal cancer with 86% accuracy. We also explored the biological relationships of the attributes used in our model to understand their relationship with cancer proliferation or suppression pathways. CONCLUSION: Our study demonstrates that the proposed model and an inexpensive miRNA testing strategy have the potential to serve as an additional method for diagnosing laryngeal cancer.

Search of inhibitors of aldose reductase for treatment of diabetic cataracts using machine learning.

Purpose: Patients with diabetes mellitus have an elevated chance of developing cataracts, a degenerative vision-impairing condition often needing surgery. The process of the reduction of glucose to sorbitol in the lens of the human eye that causes cataracts is managed by the Aldose Reductase Enzyme (AR), and it is been found that AR inhibitors may mitigate the onset of diabetic cataracts. There exists a large pool of natural and synthetic AR inhibitors that can prevent diabetic complications, and the development of a machine-learning (ML) prediction model may bring new AR inhibitors with better characteristics into clinical use. Methods: Using known AR inhibitors and their chemical-physical descriptors we created the ML model for prediction of new AR inhibitors. The predicted inhibitors were tested by computational docking to the binding site of AR. Results: Using cross-validation in order to find the most accurate ML model, we ended with final cross-validation accuracy of 90%. Computational docking testing of the predicted inhibitors gave a high level of correlation between the ML prediction score and binding free energy. Conclusions: Currently known AR inhibitors are not used yet for patients for several reasons. We think that new predicted AR inhibitors have the potential to possess more favorable characteristics to be successfully implemented after clinical testing. Exploring new inhibitors can improve patient well-being and lower surgical complications all while decreasing long-term medical expenses.

Two-Stage Deep-Learning Classifier for Diagnostics of Lung Cancer Using Metabolites.

We developed a machine-learning system for the selective diagnostics of adenocarcinoma (AD), squamous cell carcinoma (SQ), and small-cell carcinoma lung (SC) cancers based on their metabolomic profiles. The system is organized as two-stage binary classifiers. The best accuracy for classification is 92%. We used the biomarkers sets that contain mostly metabolites related to cancer development. Compared to traditional methods, which exclude hierarchical classification, our method splits a challenging multiclass task into smaller tasks. This allows a two-stage classifier, which is more accurate in the scenario of lung cancer classification. Compared to traditional methods, such a "divide and conquer strategy" gives much more accurate and explainable results. Such methods, including our algorithm, allow for the systematic tracking of each computational step.

Synergism in actions of HBV with aflatoxin in cancer development.

Aflatoxin B1 (AFB1) is a fungal metabolite found in animal feeds and human foods. It is one of the most toxic and carcinogenic of aflatoxins and is classified as a Group 1 carcinogen. Dietary exposure to AFB1 and infection with chronic Hepatitis B Virus (HBV) make up two of the major risk factors for hepatocellular carcinoma (HCC). These two major risk factors raise the probability of synergism between the two agents. This review proposes some collaborative molecular mechanisms underlying the interaction between AFB1 and HBV in accelerating or magnifying the effects of HCC. The HBx viral protein is one of the main viral proteins of HBV and has many carcinogenic qualities that are involved with HCC. AFB1, when metabolized by CYP450, becomes AFB1-exo-8,9-epoxide (AFBO), an extremely toxic compound that can form adducts in DNA sequences and induce mutations. With possible synergisms that exist between HBV and AFB1 in mind, it is best to treat both agents simultaneously to reduce the risk by HCC.

Biological functions of the Olig gene family in brain cancer and therapeutic targeting.

The Olig genes encode members of the basic helix-loop-helix (bHLH) family of transcription factors. Olig1, Olig2, and Olig3 are expressed in both the developing and mature central nervous system (CNS) and regulate cellular specification and differentiation. Over the past decade extensive studies have established functional roles of Olig1 and Olig2 in development as well as in cancer. Olig2 overexpression drives glioma proliferation and resistance to radiation and chemotherapy. In this review, we summarize the biological functions of the Olig family in brain cancer and how targeting Olig family genes may have therapeutic benefit.

Small Molecule Decoys of Aggregation for Elimination of Aß-Peptide Toxicity.

Several lines of evidence suggest that a characteristic of the neuropathology of Alzheimer's disease (AD) is the aggregation of the amyloid beta peptides (Aß), fragments of the human amyloid precursor protein (hAPP). The dominating species are the Aß40 and Aß42 fragments with 40 and 42 amino acids, respectively. Aß initially forms soluble oligomers that continue to expand to protofibrils, suggestively the neurotoxic intermediates, and thereafter turn into insoluble fibrils that are markers of the disease. Using the powerful tool of pharmacophore simulation, we selected small molecules not known to possess central nervous system (CNS) activity but that might interact with Aß aggregation, from the NCI Chemotherapeutic Agents Repository, Bethesda, MD. We assessed the activity of these compounds on Aß aggregation using the thioflavin T fluorescence correlation spectroscopy (ThT-FCS) assay. Förster resonance energy transfer-based fluorescence correlation spectroscopy (FRET-FCS) was used to characterize the dose-dependent activity of selected compounds at an early stage of Aß aggregation. Transmission electron microscopy (TEM) confirmed that the interfering substances block fibril formation and identified the macrostructures of Aß aggregates formed in their presence. We first found three compounds generating protofibrils with branching and budding never observed in the control. One compound generated a two-dimensional sheet structure and another generated a double-stranded filament. Importantly, these compounds generating protofibrils with altered macrostructure protected against Aß-induced toxicity in a cell model while showing no toxicity in a model of cognition in normal mice. The data suggest that the active compounds act as decoys turning the aggregation into nontoxic trajectories and pointing toward novel approaches to therapy.

Diagnostics of ovarian cancer via metabolite analysis and machine learning.

Ovarian cancer (OC) is the second most common cancer of the female reproductive system. Due to the asymptomatic nature of early stages of OC and an increasingly poor prognosis in later stages, methods of screening for OC are much desired. Furthermore, screening and diagnosis processes, in order to justify use on asymptomatic patients, must be convenient and non-invasive. Recent developments in machine-learning technologies have made this possible via techniques in the field of metabolomics. The objective of this research was to use existing metabolomics data on OC and various analytic methods to develop a machine-learning model for the classification of potentially OC-related metabolite biomarkers. Pathway analysis and metabolite-set enrichment analysis were performed on gathered metabolite sets. Quantitative molecular descriptors were then used with various machine-learning classifiers for the diagnostics of OC using related metabolites. We elucidated that the metabolites associated with OC used for machine-learning models are involved in five metabolic pathways linked to OC: Nicotinate and Nicotinamide Metabolism, Glycolysis/Gluconeogenesis, Aminoacyl-tRNA Biosynthesis, Valine, Leucine and Isoleucine Biosynthesis, and Alanine, Aspartate and Glutamate Metabolism. Several classification models for the identification of OC using related metabolites were created and their accuracies were confirmed through testing with 10-fold cross-validation. The most accurate model was able to achieve 85.29% accuracy. The elucidation of biological pathways specific to OC using metabolic data and the observation of changes in these pathways in patients have the potential to contribute to the development of screening techniques for OC. Our results demonstrate the possibility of development of the machine-learning models for OC diagnostics using metabolomics data.

Repurposing of drugs for combined treatment of COVID-19 cytokine storm using machine learning.

Severe acute respiratory syndrome coronavirus 2 (SARS­CoV­2) induced cytokine storm is the major cause of COVID-19 related deaths. Patients have been treated with drugs that work by inhibiting a specific protein partly responsible for the cytokines production. This approach provided very limited success, since there are multiple proteins involved in the complex cell signaling disease mechanisms. We targeted five proteins: Angiotensin II receptor type 1 (AT1R), A disintegrin and metalloprotease 17 (ADAM17), Nuclear Factor­Kappa B (NF­κB), Janus kinase 1 (JAK1) and Signal Transducer and Activator of Transcription 3 (STAT3), which are involved in the SARS­CoV­2 induced cytokine storm pathway. We developed machine-learning (ML) models for these five proteins, using known active inhibitors. After developing the model for each of these proteins, FDA-approved drugs were screened to find novel therapeutics for COVID­19. We identified twenty drugs that are active for four proteins with predicted scores greater than 0.8 and eight drugs active for all five proteins with predicted scores over 0.85. Mitomycin C is the most active drug across all five proteins with an average prediction score of 0.886. For further validation of these results, we used the PyRx software to conduct protein-ligand docking experiments and calculated the binding affinity. The docking results support findings by the ML model. This research study predicted that several drugs can target multiple proteins simultaneously in cytokine storm-related pathway. These may be useful drugs to treat patients because these therapies can fight cytokine storm caused by the virus at multiple points of inhibition, leading to synergistically effective treatments.

Diagnostics of Thyroid Cancer Using Machine Learning and Metabolomics.

The objective of this research is, with the analysis of existing data of thyroid cancer (TC) metabolites, to develop a machine-learning model that can diagnose TC using metabolite biomarkers. Through data mining, pathway analysis, and machine learning (ML), the model was developed. We identified seven metabolic pathways related to TC: Pyrimidine metabolism, Tyrosine metabolism, Glycine, serine, and threonine metabolism, Pantothenate and CoA biosynthesis, Arginine biosynthesis, Phenylalanine metabolism, and Phenylalanine, tyrosine, and tryptophan biosynthesis. The ML classifications' accuracies were confirmed through 10-fold cross validation, and the most accurate classification was 87.30%. The metabolic pathways identified in relation to TC and the changes within such pathways can contribute to more pattern recognition for diagnostics of TC patients and assistance with TC screening. With independent testing, the model's accuracy for other unique TC metabolites was 92.31%. The results also point to a possibility for the development of using ML methods for TC diagnostics and further applications of ML in general cancer-related metabolite analysis.

Modeling of mutant superoxide dismutase 1 octamers with cross-linked disulfide bonds.

Mutant superoxide dismutase 1 (SOD1) may form cyclic structures due to its greater instability from aberrant demetallization and oxidation of cysteine bonds. This cyclic structure may allow SOD1 to form ion channels on membranes such as the mitochondrial membrane, causing imbalances in the concentration of intracellular ions as a potential mechanism for the progressive neuron death involved in amyotrophic lateral sclerosis (ALS). Using docking programs within modeling software, models of mutant SOD1 dimers and eventually ring oligomers were constructed based on known descriptions of such structures in addition to information on the orientation of the models associated with a membrane. The resulting structure consists of a ring of four demetallated mutant SOD1 dimers with cross-linked disulfide bonds. Stability of the octamer model was supported by the molecular dynamics simulations. Further analysis of the octamer model indicated that its inner- and outer-pore diameters were stable, matching the dimensions of known SOD1 ion channels.

Interactions of large T-Antigen (LT) protein of polyomaviruses with p53 unfold their cancerogenic potential.

Polyomaviruses such as Simian Virus 40 (SV40) and John Cunningham Virus (JCV) have been extensively studied for their potential role in aiding oncogenic transformation. One of the mechanisms through which they do this is by inactivating p53, a known tumor suppressor, through one of their viral proteins, large T-antigen (LT). However, these two viruses represent only a fraction of existing polyomaviruses. Using Clustal Omega, we aligned the protein sequences of LT for 12 different polyomaviruses and found high similarity across polyomavirus LT. We then utilized Molecular Operating Environment (MOE) v2019.01 to compare the binding of SV40 LT to p53 and p53 to DNA to more precisely define the mechanism with which SV40 LT inactivates p53. By binding to p53 residues essential to DNA binding, SV40 LT prevents the proper interaction of p53 with DNA and consequently its fulfillment of transcription factor functions. To further explore the possibility for other polyomavirus LT to do the same, we either retrieved existing 3D structures from RCSB Protein Data Bank or generated 3D homology models of other polyomavirus LT and modeled their interactions with p53. These models interacted with p53 in a similar manner as SV40 LT and provide further evidence of the potential of other polyomavirus LT to inactivate p53. This work demonstrates the importance of investigating the oncogenic potential of polyomaviruses and elucidates future targets for cancer treatment.Communicated by Ramaswamy H. Sarma.

Implications of viral infection in cancer development.

Since the identification of the first human oncogenic virus in 1964, viruses have been studied for their potential role in aiding the development of cancer. Through the modulation of cellular pathways associated with proliferation, immortalization, and inflammation, viral proteins can mimic the effect of driver mutations and contribute to transformation. Aside from the modulation of signaling pathways, the insertion of viral DNA into the host genome and the deregulation of cellular miRNAs represent two additional mechanisms implicated in viral oncogenesis. In this review, we will discuss the role of twelve different viruses on cancer development and how these viruses utilize the abovementioned mechanisms to influence oncogenesis. The identification of specific mechanisms behind viral transformation of human cells could further elucidate the process behind cancer development.

Finding distinctions between oral cancer and periodontitis using saliva metabolites and machine learning.

OBJECTIVE: The aim of this research is the study of metabolic pathways related to oral cancer and periodontitis along with development of machine-learning model for elucidation of these diseases based on saliva metabolites of patients. METHODS: Data mining, metabolomic pathways analysis, study of metabolite-gene networks related to these diseases. Machine-learning and deep-learning methods for development of the model for recognition of oral cancer versus periodontitis, using patients' saliva. RESULTS: The most accurate classifications between oral cancer and periodontitis were performed using neural networks, logistic regression and stochastic gradient descent confirmed by the separate 10-fold cross-validations. The best results were achieved by the deep-learning neural network with the TensorFlow program. Accuracy of the resulting model was 79.54%. The other methods, which did not rely on deep learning, were able to achieve comparable, although slightly worse results with respect to accuracy. CONCLUSION: Our results demonstrate a possibility to distinguish oral cancer from periodontal disease by analysis the saliva metabolites of a patient, using machine-learning methods. These findings may be useful in the development of a non-invasive method to aid care providers in determining between oral cancer and periodontitis quickly and effectively.

Machine learning model to predict oncologic outcomes for drugs in randomized clinical trials.

Predicting oncologic outcome is challenging due to the diversity of cancer histologies and the complex network of underlying biological factors. In this study, we determine whether machine learning (ML) can extract meaningful associations between oncologic outcome and clinical trial, drug-related biomarker and molecular profile information. We analyzed therapeutic clinical trials corresponding to 1102 oncologic outcomes from 104 758 cancer patients with advanced colorectal adenocarcinoma, pancreatic adenocarcinoma, melanoma and nonsmall-cell lung cancer. For each intervention arm, a dataset with the following attributes was curated: line of treatment, the number of cytotoxic chemotherapies, small-molecule inhibitors, or monoclonal antibody agents, drug class, molecular alteration status of the clinical arm's population, cancer type, probability of drug sensitivity (PDS) (integrating the status of genomic, transcriptomic and proteomic biomarkers in the population of interest) and outcome. A total of 467 progression-free survival (PFS) and 369 overall survival (OS) data points were used as training sets to build our ML (random forest) model. Cross-validation sets were used for PFS and OS, obtaining correlation coefficients (r) of 0.82 and 0.70, respectively (outcome vs model's parameters). A total of 156 PFS and 110 OS data points were used as test sets. The Spearman correlation (rs ) between predicted and actual outcomes was statistically significant (PFS: rs = 0.879, OS: rs = 0.878, P < .0001). The better outcome arm was predicted in 81% (PFS: N = 59/73, z = 5.24, P < .0001) and 71% (OS: N = 37/52, z = 2.91, P = .004) of randomized trials. The success of our algorithm to predict clinical outcome may be exploitable as a model to optimize clinical trial design with pharmaceutical agents.

Prediction of Premature Termination Codon Suppressing Compounds for Treatment of Duchenne Muscular Dystrophy Using Machine Learning.

A significant percentage of Duchenne muscular dystrophy (DMD) cases are caused by premature termination codon (PTC) mutations in the dystrophin gene, leading to the production of a truncated, non-functional dystrophin polypeptide. PTC-suppressing compounds (PTCSC) have been developed in order to restore protein translation by allowing the incorporation of an amino acid in place of a stop codon. However, limitations exist in terms of efficacy and toxicity. To identify new compounds that have PTC-suppressing ability, we selected and clustered existing PTCSC, allowing for the construction of a common pharmacophore model. Machine learning (ML) and deep learning (DL) models were developed for prediction of new PTCSC based on known compounds. We conducted a search of the NCI compounds database using the pharmacophore-based model and a search of the DrugBank database using pharmacophore-based, ML and DL models. Sixteen drug compounds were selected as a consensus of pharmacophore-based, ML, and DL searches. Our results suggest notable correspondence of the pharmacophore-based, ML, and DL models in prediction of new PTC-suppressing compounds.

Bovine leukemia virus relation to human breast cancer: Meta-analysis.

Bovine leukemia virus (BLV) is a virus that infects cattle around the world and is very similar to the human T-cell leukemia virus (HTLV), which causes adult T-cell leukemia/lymphoma (ATL). Recently, presence of BLV DNA and protein was demonstrated in commercial bovine products and in humans. BLV DNA is generally found at higher rates in humans who have or will develop breast cancer, according to research done with subjects from several countries. These findings have led to a hypothesis that BLV transmission plays a role in breast cancer oncogenesis in humans. Here we summarize the current knowledge in the field.

A strategy for designing allosteric modulators of transcription factor dimerization.

Transcription factors (TFs) are fundamental in the regulation of gene expression in the development and differentiation of cells. They may act as oncogenes and when overexpressed in tumors become plausible targets for the design of antitumor agents. Homodimerization or heterodimerization of TFs are required for DNA binding and the association interface between subunits, for the design of allosteric modulators, appears as a privileged structure for the pharmacophore-based computational strategy. Based on this strategy, a set of compounds were earlier identified as potential suppressors of OLIG2 dimerization and found to inhibit tumor growth in a mouse glioblastoma cell line and in a whole-animal study. To investigate whether the antitumor activity is due to the predicted mechanism of action, we undertook a study of OLIG2 dimerization using fluorescence cross-correlation spectroscopy (FCCS) of live HEK cells transfected with 2 spectrally different OLIG2 clones. The selected compounds showed an effect with potency, which correlated with the earlier observed antitumor activity. The OLIG2 proteins showed change in diffusion time under compound treatment in line with dissociation from DNA. The data suggest a general approach of drug discovery based on the design of allosteric modulators of protein-protein interaction.

ANKRD9 is a metabolically-controlled regulator of IMPDH2 abundance and macro-assembly.

Members of a large family of Ankyrin Repeat Domain (ANKRD) proteins regulate numerous cellular processes by binding to specific protein targets and modulating their activity, stability, and other properties. The same ANKRD protein may interact with different targets and regulate distinct cellular pathways. The mechanisms responsible for switches in the ANKRDs' behavior are often unknown. We show that cells' metabolic state can markedly alter interactions of an ANKRD protein with its target and the functional outcomes of this interaction. ANKRD9 facilitates degradation of inosine monophosphate dehydrogenase 2 (IMPDH2), the rate-limiting enzyme in GTP biosynthesis. Under basal conditions ANKRD9 is largely segregated from the cytosolic IMPDH2 in vesicle-like structures. Upon nutrient limitation, ANKRD9 loses its vesicular pattern and assembles with IMPDH2 into rodlike filaments, in which IMPDH2 is stable. Inhibition of IMPDH2 activity with ribavirin favors ANKRD9 binding to IMPDH2 rods. The formation of ANKRD9/IMPDH2 rods is reversed by guanosine, which restores ANKRD9 associations with the vesicle-like structures. The conserved Cys109Cys110 motif in ANKRD9 is required for the vesicle-to-rods transition as well as binding and regulation of IMPDH2. Oppositely to overexpression, ANKRD9 knockdown increases IMPDH2 levels and prevents formation of IMPDH2 rods upon nutrient limitation. Taken together, the results suggest that a guanosine-dependent metabolic switch determines the mode of ANKRD9 action toward IMPDH2.