Gene Expression Profiling to detect Survival Prediction of Diffuse Large B-Cell Lymphoma

Shreelekha Pore, National Institute of Technology, Rourkela

25% of non-Hodgkin Lymphoma includes Diffuse Large B-Cell Lymphoma (DLBCL). Through the study of cell-of-origin (COO) using gene expression profiling (GEP), DLBCL has been classified as- germinal centre B-cell-like and activated B-cell-like. Artificial intelligence is widely used for prognostic purposes for certain diseases. Machine Learning can be used for prediction of outcomes through the study of various complex interactions among many variables. Firstly, the transcripts and patterns of gene expression linked to prognosis were studied. Then the information was used to identify random forest model for detection of overall survival possibilities in DLBCL. 

Gene expression databases- GSE10846 and GSE23501 were studied. GSE10846 showed data obtained from whole tissue biopsies of 420 DLBCL patients. 233 had undergone R-CHOP treatment. GSE23501 on the other hand showed 69 DLBCL patients with whole tissue biopsies who had undergone R-CHOP- like a regimen. Spearman correlation was used to nullify duplicate samples and finally set 64 cases were validated. Hence gene expression data inferred COO classification. GEO was used to obtain expression data which was Log-2 transformed for both cohorts. Mclust algorithm was used to study for cluster prediction and verified by cox-regression. Shoenfeld’s test helped to assume proportional hazards. Random forest survival models made with rfsrc function is applied in randomForestSRC package in R. To optimize mtry and nnodes variables, parameter tuning was performed by tune.rfscr function. 

Continuous Rank Probability score (CRPS) with 0 as bets result. Predict.rfscrc function was used for survival prediction. Harrel’s concordance index (C-index) below 0.5 means poor prediction accuracy, near 0.5 means random guessing and 1 means perfect predictions. Thus the best model in terms of C-index was chosen for replication in the validation set. Probes corresponding to genes TNFRSF9 and BCL2L1 with p-values 0.04 and 8.59X10^-3 were involved in survival tests. Cluster prediction resulted in the classification of a group of 20.31% of patients. Multivariate cox-regression showed association with dreadful outcome (p-value 5.43X10⁺³, HR 6.80, 95% CIHR 1.76-26.26). GEP_0.01, GEP_0.05, GEP_0.1 gave survival predictions with GEP_0.1 being best (102 genes) with training C-index 0.7783 and test C-index 0.7415. However, the best model emerged as GEP_0.1 (4 genes) clusterization and COO classification with training and test C-index as 0.8051 and 0.7615. Expression of MS4A4A was highest. Expression of SLIT2, NEAT1, CPT1A, IGSF9, CD302 was superior to COO classification. We can conclude that applying machine learning algorithms to gene expression and clinical data is useful to detect DLBCL survival models.

Also read: Discovery of a better gene-editing tool?

Reference: Mosquera Orgueira, A, et al. “Improved personalized survival prediction of patients with diffuse large B-cell lymphoma using gene expression profiling.” BMC Cancer 20, Oct 2020. 

DOI: 10.21203/rs.3.rs-40793/v1

https://www.researchgate.net/publication/342855418_Improved_Personalized_Survival_Prediction_of_Patients_With_Diffuse_Large_B-cell_Lymphoma_Using_Gene_Expression_Profiling

• 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