Involvement Of Bayesian Network Models In Predicting Various Types Of Hematological Malignancies

 Acute Myeloid Leukemia (AML) is a kind of myeloid blood malignancy in which the bone marrow produces aberrant white blood cells, red blood cells, or platelets. It is the most predominant acute leukaemia in adults, and it predominantly affects the elderly. It is a deadly kind of blood cancer that causes around 1.2 % of all cancer fatalities in the United States.

Myelodysplastic Syndrome (MDS) is bone marrow and a blood disorder that damages myeloid cells. Abnormal hematopoiesis, or the inefficient generation of blood cells and platelets in the bone marrow, is a hallmark of MDS. MDS, unlike AML, is generally benign and has a low mortality risk, but it can advance over time, with 30% of MDS patients progressing to AML. As a result, it’s critical to compare these two illnesses and give scientific insights into their molecular parallels and variances. For detecting small but coordinated changes in expression of an interacting and linked group of genes, network analysis is the best method. The AML and MDS were classified using a new technique based on coexpression networks and Bayesian networks. The approach is showed schematically in Figure 1.

In AML, WGCNA was used to organise related genes into gene modules (clusters) based on their coexpression patterns. WGCNA uses the average linkage hierarchical method to cluster the genes6. WGCNA calculates one eigen gene for each gene module, which summarises the biological information in that module into a single value per sample.

These eigen genes were used to train a Bayesian network (BN) with nodes (random variables) representing gene modules and directed edges (arcs) representing conditional dependencies between the eigen genes. Gene expression data and gene regulatory networks have both been modelled using Bayesian networks. A directed acyclic graph (DAG) and its accompanying conditional probability density functions make up a BN.

Significance of network analysis:

Biological activities in cells frequently need the cooperation of several genes. Network analysis can discover small but coordinated changes in a group of genes that connect and have similar functions. As a result, network analysis outperforms traditional techniques based on a list of differentially expressed genes. A coexpression network, in particular, stimulates the interaction between several genes based on their coexpression pattern. The eigen genes that arise, which summarise the biological information of modules, are noise and profiling platform resistant. This was proven by comparing eigengenes efficacy and differentially expressed genes in support vector machine learning. We coupled coexpression network analysis with Bayesian networks to describe the interactions between hundreds of genes in one network.

Conclusion

The use of network analysis to derive relevant biomarkers (features) from gene expression patterns is beneficial. Eigengenes, in particular, have more predictive potential than individual genes. These data may be used to construct a Bayesian network to describe the relationship between the gene modules and the biological or clinical state of interest. (SVM), demonstrating that the power of our method is based on how we use eigengenes as biological fingerprints (i.e., features). When individual genes are used as features, the SVM performs poorly, but when eigengenes are used as features, it performs similarly to the Bayesian network. Nonetheless, unlike SVM, which is more of a black box classifier, our Bayesian network technique is beneficial since it quickly delineates the most related characteristics with the illness type.