Pfeature: A Tool for Computing Wide Range of Protein Features and Building Prediction Models

In the last three decades, a wide range of protein features have been discovered to annotate a protein. Numerous attempts have been made to integrate these features in a software package/platform so that the user may compute a wide range of features from a single source. To complement the existing methods, we developed a method, Pfeature, for computing a wide range of protein features. Pfeature allows to compute more than 200,000 features required for predicting the overall function of a protein, residue-level annotation of a protein, and function of chemically modified peptides. It has six major modules, namely, composition, binary profiles, evolutionary information, structural features, patterns, and model building. Composition module facilitates to compute most of the existing compositional features, plus novel features. The binary profile of amino acid sequences allows to compute the fraction of each type of residue as well as its position. The evolutionary information module allows to compute evolutionary information of a protein in the form of a position-specific scoring matrix profile generated using Position-Specific Iterative Basic Local Alignment Search Tool (PSI-BLAST); fit for annotation of a protein and its residues. A structural module was developed for computing of structural features/descriptors from a tertiary structure of a protein. These features are suitable to predict the therapeutic potential of a protein containing non-natural or chemically modified residues. The model-building module allows to implement various machine learning techniques for developing classification and regression models as well as feature selection. Pfeature also allows the generation of overlapping patterns and features from a protein. A user-friendly Pfeature is available as a web server python library and stand-alone package.

Computational resources for identification of cancer biomarkers from omics data

Cancer is one of the most prevailing, deadly and challenging diseases worldwide. The advancement in technology led to the generation of different types of omics data at each genome level that may potentially improve the current status of cancer patients. These data have tremendous applications in managing cancer effectively with improved outcome in patients. This review summarizes the various computational resources and tools housing several types of omics data related to cancer. Major categorization of resources includes-cancer-associated multiomics data repositories, visualization/analysis tools for omics data, machine learning-based diagnostic, prognostic, and predictive biomarker tools, and data analysis algorithms employing the multiomics data. The review primarily focuses on providing comprehensive information on the open-source multiomics tools and data repositories, owing to their broader applicability, economic-benefit and usability. Sections including the comparative analysis, tools applicability and possible future directions have also been discussed in detail. We hope that this information will significantly benefit the researchers and clinicians, especially those with no sound background in bioinformatics and who lack sufficient data analysis skills to interpret something from the plethora of cancer-specific data generated nowadays.

In-silico identification of subunit vaccine candidates against lung cancer-associated oncogenic viruses

Globally, ~20% of cancer malignancies are associated with virus infections. Lung cancer is the most prevalent cancer and has a 10% 5-year survival rate when diagnosed at stage IV. Cancer vaccines and oncolytic immunotherapy are promising treatment strategies for better clinical outcomes in advanced-stage cancer patients. Here, we used a reverse vaccinology approach to devise subunit vaccine candidates against lung cancer-causing oncogenic viruses. Protein components (945) from nine oncogenic virus species were systematically analyzed to identify epitope-based subunit vaccine candidates. Best vaccine candidates were identified based on their predicted ability to stimulate humoral and cell-mediated immunity and avoid self-tolerance. Using a rigorous integrative approach, we identified 125 best antigenic epitopes with predicted B-cell, T-cell, and/or MHC-binding capability and vaccine adjuvant potential. Thirty-two of these antigenic epitopes were predicted to have IL-4/IFN-gamma inducing potential and IL-10 non-inducing potential and were predicted to bind 15 MHC-type I and 49 MHC-type II alleles. All 32 epitopes were non-allergenic and 31 were non-toxic. The identified epitopes showed good conservancy and likely bind a broad class of human HLA alleles, indicating promiscuous potential. The majority of best antigenic epitopes were derived from Human papillomavirus and Epstein-Barr virus proteins. Of the 32 epitopes, 25 promiscuous epitopes were related to E1 and E6 envelope genes and were present in multiple viral strains/species, potentially providing heterologous immunity. Further validating our results, 38 antigenic epitopes were also present in the largest experimentally-validated epitope resource, Immune Epitope Database and Analysis Resource. We further narrowed the selection to 29 antigenic epitopes with the highest immunogenic/immune-boosting potential. These epitopes possess tremendous therapeutic potential as vaccines against lung cancer-causing viruses and should be validated in future experiments. All findings are available at https://webs.iiitd.edu.in/raghava/vlcvirus/.

Risk prediction in cutaneous melanoma patients from their clinico-pathological features: superiority of clinical data over gene expression data

Risk assessment in cutaneous melanoma (CM) patients is one of the major challenges in the effective treatment of CM patients. Traditionally, clinico-pathological features such as Breslow thickness, American Joint Committee on Cancer (AJCC) tumor staging, etc. are utilized for this purpose. However, due to advancements in technology, most of the upcoming risk prediction methods are gene-expression profile (GEP) based. In this study, we have tried to develop new GEP and clinico-pathological features-based biomarkers and assessed their prognostic strength in contrast to existing prognostic methods. We developed risk prediction models using the expression of the genes associated with different cancer-related pathways and got a maximum hazard ratio (HR) of 2.52 with p-value ~10-8 for the apoptotic pathway. Another model, based on combination of apoptotic and notch pathway genes boosted the HR to 2.57. Next, we developed models based on individual clinical features and got a maximum HR of 2.45 with p-value ~10-6 for Breslow thickness. We also developed models using the best features of clinical as well as gene-expression data and obtained a maximum HR of 3.19 with p-value ~10-9. Finally, we developed a new ensemble method using clinical variables only and got a maximum HR of 6.40 with p-value ~10-15. Based on this method, a web-based service and an android application named 'CMcrpred' is available at (https://webs.iiitd.edu.in/raghava/cmcrpred/) and Google Play Store respectively to facilitate scientific community. This study reveals that our new ensemble method based on only clinico-pathological features overperforms methods based on GEP based profiles as well as currently used AJCC staging. It also highlights the need to explore the full potential of clinical variables for prognostication of cancer patients.

Identification of prognostic biomarkers for major subtypes of non-small-cell lung cancer using genomic and clinical data

PURPOSE

Intra-tumor heterogeneity and high mortality among patients with non-small-cell lung carcinoma (NSCLC) emphasize the need to identify reliable prognostic markers unique to each subtype.

METHODS

In this study, univariate cox regression and prognostic index (PI)-based approaches were used to develop models for predicting NSCLC patients' subtype-specific survival.

RESULTS

Prognostic analysis of TCGA dataset identified 1334 and 2129 survival-specific genes for LUSC (488 samples) and LUAD (497 samples), respectively. Individually, 32 and 271 prognostic genes were found and validated in GSE study exclusively for LUSC and LUAD. Nearly, 9-10% of the validated genes in each subtype were already reported in multiple studies thus highlighting their importance as prognostic biomarkers. Strong literature evidence against these prognostic genes like "ELANE" (LUSC) and "AHSG" (LUAD) instigates further investigation for their therapeutic and diagnostic roles in the corresponding cohorts. Prognostic models built on five and four genes were validated for LUSC [HR = 2.10, p value = 1.86 × 10^-5] and LUAD [HR = 2.70, p value = 3.31 × 10^-7], respectively. The model based on the combination of age and tumor stage performed well in both NSCLC subtypes, suggesting that despite having distinctive histological features and treatment paradigms, some clinical features can be good prognostic predictors in both.

CONCLUSION

This study advocates that investigating the survival-specific biomarkers restricted to respective cohorts can advance subtype-specific prognosis, diagnosis, and treatment for NSCLC patients. Prognostic models and markers described for each subtype may provide insight into the heterogeneity of disease etiology and help in the development of new therapeutic approaches for the treatment of NSCLC patients.

Prediction of risk scores for colorectal cancer patients from the concentration of proteins involved in mitochondrial apoptotic pathway

One of the major challenges in managing the treatment of colorectal cancer (CRC) patients is to predict risk scores or level of risk for CRC patients. In past, several biomarkers, based on concentration of proteins involved in type-2/intrinsic/mitochondrial apoptotic pathway, have been identified for prognosis of colorectal cancer patients. Recently, a prognostic tool DR_MOMP has been developed that can discriminate high and low risk CRC patients with reasonably high accuracy (Hazard Ratio, HR = 5.24 and p-value = 0.0031). This prognostic tool showed an accuracy of 59.7% when used to predict favorable/unfavorable survival outcomes. In this study, we developed knowledge based models for predicting risk scores of CRC patients. Models were trained and evaluated on 134 stage III CRC patients. Firstly, we developed multiple linear regression based models using different techniques and achieved a maximum HR value of 6.34 with p-value = 0.0032 for a model developed using LassoLars technique. Secondly, models were developed using a parameter optimization technique and achieved a maximum HR value of 38.13 with p-value 0.0006. We also predicted favorable/unfavorable survival outcomes and achieved maximum prediction accuracy value of 71.64%. A further enhancement in the performance was observed if clinical factors are added to this model. Addition of age as a variable to the model improved the HR to 40.11 with p-value as 0.0003 and also boosted the accuracy to 73.13%. The performance of our models were evaluated using five-fold cross-validation technique. For providing service to the community we also developed a web server ‘CRCRpred’, to predict risk scores of CRC patients, which is freely available at https://webs.iiitd.edu.in/raghava/crcrpred.