Understanding mutation data generated in biomedical research stimulates the development of computational methods. Previous studies have revealed structural and functional impacts on underlying proteins from variants, and research has proven that these impacts can differentiate between disease-associated and neutral mutations. Most current prediction tools have taken advantage of these characteristics, along with evolutionary information readily available from sequence alignment. Such tools have demonstrated impressive classification

MutPred Score

Loss of methylation

Gain of ubiquitination

Loss of catalytic residue

Gain of MoRF binding

Gain of catalytic residue

Gain of methylation

Loss of MoRF binding

MutPred Score

Figure 6. The distribution of MutPred scores for nsSNPs from dbSNP (left), and the top ten hypotheses for disease-associated mutations (right). The density on the left is a normalized frequency to ensure a total area in the bar plot equals one.

accuracy in monogenic disease-associated mutations but have performed less well for cancer somatic mutations. One explanation from an evolutionary perspective for this descrepency is that cancers usually arise late in life, so they are subjected to less purifying selection. This makes conservation information in cancers less useful than in monogenic diseases [56]. This field faces two immediate challenges: (1) How can we improve these tools to improve performance with somatic mutations? If the consensus opinion holds that tools depending on evolutionary knowledge are less effective than when applied to monogenic-disease-related mutations, it seems that research should explore other avenues. Inclusion of the mutation context in the model-e.g., pathways containing disease proteins-might offer a starting point for new directions. (2) How can we more accurately elucidate the molecular mechanisms for predicted deleterious mutations? MutPred has demonstrated this concept through definitions of gain/loss of individual properties. Similar features should be considered once they prove capable of reliably discriminating between disease-associated and neutral mutations. By continuously improving our computational tools, we can obtain better and more accurate understandings of biology and human health.

Loss of disorder

Gain of disorder

Was this article helpful?

0 0

Post a comment