neoDL: a model that recognizes longest survival glioblastomas

Parnad Basu, Amity University Kolkata

What is glioblastoma?

When the cells of our body start to grow at a rapid speed and uncontrollably, a tumor is formed. A tumor can be both dangerous (malignant) and not dangerous (benign). In the case of malignant tumors, it tends to spread from one body part to the other rapidly, which causes cancer. Cancer can start at any place in our body and can spread anywhere. Glioblastoma is a type of brain cancer that is very aggressive. Aggressive means it grows fast and spreads quickly. Glioblastoma is formed from star-shaped cells in the brain known as astrocytes. Although glioblastomas are not that common, it is very deadly.

Over the years, the survival rate of glioblastoma has changed because of therapies like radiation, chemotherapy, targeted therapy, etc. However, the prognosis of glioblastomas is generally poor and depends upon a single individual. Which brings down the average survival time to merely 14 months. This brings us to the current study done on glioblastomas.

What is the recent study all about?

This recent study, published in 2021, talks about how neoantigen can provide better results in the case of glioblastomas. Neoantigens are a new protein that is being formed when certain mutations occur in a malignant tumor DNA. It helps fight cancer by creating an immune response. Personalized neoantigen-based immune therapies have shown promising results in the case of melanoma and lung cancer already. In this study, scientists formulated neoDL which is a novel neoantigen intrinsic feature-based deep learning model.

IDH (isocitrate dehydrogenase) are gene mutations, that can act as a prognostic marker to diffuse glioblastomas. The model created by scientists can be therapeutically manipulated to identify IDH wild-type glioblastomas. This study was done to successfully distinguish IDH wild-type glioblastomas into different prognostic subgroups. Doing so helps to identify those who would benefit from neoantigen-based personalized immunotherapies. Neoantigens enable tumor-specific T-cell responses that cause antitumor immune responses in return. However, the number of high-quality neoantigens is very less which makes the clinical application difficult.

Methods used to conduct the study:

1. Data description- Major information was collected from the ATLAS-TCGA pan-glioma study, TCGA Data portal, Pri cohort.
2. Neoantigen feature calculation- 2928 features from 2263 neoantigens were extracted.
3. Prognostic feature selection- For all neoantigens and wild-type peptides all features were calculated and an average for each case was done.
4. k-means clustering- Hierarchical k-means clustering was done to separate two clusters.
5. Deep learning model- Valid features of cohorts were used to prepare the deep learning model.
6. LOOCV (Leave one out cross-validation)- Cohorts were separated randomly between training and test sets at the ratio of 6 to 4.

The outcome of the study:

The results achieved by using the neoDL model are significantly better than any pre-existing models. Along with two independent data cohorts (KM and TCGA), the neoantigen model achieved better predictive performances in IDH wild-type glioblastomas. In addition to that, it was found that neoDL performed better than the majority of the deep learning models (DeepLearningModel and PASNet). This model also helped to find out two correlated neoantigen features as VHSE2 and protFP2 which helped dividing glioblastomas into different subgroups.

Furthermore, neoDL also showed a difference between short- and long-term survival of IDH wild-type glioblastomas. In which the long-term survival showed a higher molecular weight of dipeptide, molecular size-related features, and electrostatic potential-related features. Also, the neoDL model was capable to divide patients into two distinct subgroups for glioblastoma, IDH wildtype, Classical-like, Classical,  and Mesenchymal-like subtypes.

Conclusion:

Cancer is a life-threatening disease. New models and therapeutic techniques are much needed to fight it. In this study, only the sequence structure was the main focus. Further improvement of this model with secondary and tertiary protein structure is required. Also, some other effective deep learning methods are needed to be augmented.

Also read: Si-RNA nanoparticles used to treat neuroblastoma

Reference: Sun, T., He, Y., Li, W., Liu, G., Li, L., Wang, L., Xiao, Z., Han, X., Wen, H., Liu, Y., Chen, Y., Wang, H., Li, J., Fan, Y., Zhang, W., Zhang, J. (2021). neoDL: a novel neoantigen intrinsic feature-based deep learning model identifies IDH wild-type glioblastomas with the longest survival. BMC Bioinformatics. https://doi.org/10.1186/s12859-021-04301-6

• The Corrosion Prediction from the Corrosion Product Performance
• Nitrogen Resilience in Waterlogged Soybean plants
• Cell Senescence in Type II Diabetes: Therapeutic Potential
• Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
• AI Literacy in Early Childhood Education: Challenges and Opportunities