james gomes




Rapid computation and interpretation of Boolean attractors in biological networks

A multitude of interaction networks exist between different molecular entities within the intracellular milieu. By representing these interactions as a Boolean network, cell fate can be correlated to singleton attractors, and used to shortlist genes implicated in disease pathology. Detection of these singleton attractors is an NP-hard problem. We report here a sequential subgraph (SSG) algorithm that identifies all singleton attractors of a biological network and reduces the number of computations to do so by several orders of magnitude compared with explicit enumeration (EE) while retaining accuracies. The SSG algorithm deconstructs the biological network into subgraphs of sizes equal to their in-degrees. For each subgraph, the states constituting singleton attractors are computed separately and then stitched together according to a computed sequence to obtain the complete set. We applied this algorithm to determine the attractors of the γ -secretase network consisting of 146 vertices and 193 edges, a near impossible task by EE. For this network, we also simulated the effect of gain of function of PSEN1 observed in Alzheimer’s patients, and compared the differences in the 550 attractors with those obtained for normal PSEN1 activity. The proteins exhibiting differential activity were segregated and categorized into apoptosis, Ca2+ signalling, amyloidosis, Notch signalling, oxidative stress, MAPK cascade, cell cycle and proliferation clusters. By segregating proteins in this manner from the attractor states, it was possible to elucidate the metabolic impact of PSEN1 mutation in Alzheimer’s disease.


A comparison of performance of the SSG algorithm with the EE method for determination of singleton attractors: (a) CPU time and (b) number of computations. A thousand different random networks were created for each of sizes 5, 10, 15, 20 and 25, with constant in-degree ki =1,2,3,4 but different interactions, so that these mimicked biological networks. As the size of the network increases, for N =5 to 25, the time taken by SSG to determine the attractors becomes shorter compared with EE. For example, when N =25, the SSG algorithm takes two orders of magnitude fewer CPU time to determine the singleton attractors. The straight line for EE on the log10 scale shows that this method depends only on N, the total number of vertices. However, the SSG algorithm depends on the in-degree and the total number of vertices.


Mechanisms of Loss of Function of Angiogenin causing ALS

Missense mutations in the coding region of angiogenin (ANG) gene have been reported in patients suffering from Amyotrophic Lateral Sclerosis (ALS) - a fatal neurodegenerative disorder characterized by the selective devastation of motor neurons. Most of the ALS-causing mutations in Angiogenin affect either its ribonucleolytic or nuclear translocation activity. However, the molecular mechanisms that lead to loss of these functions are not understood. To address this, we carried out extensive Molecular Dynamics (MD) simulations (explicit and implicit) of wild-type Angiogenin and its mutants. From the MD simulations, we determine certain structural and dynamic attributes, and correlate and predict the loss-of-function(s) mechanisms. Our simulation results show that while the loss of ribonucleolytic activity of Angiogenin is due to a characteristic conformational switching of the catalytic residue His114, the loss of nuclear translocation activity is due to the local folding of nuclear localization signal residues 31RRR33, resulting in reduction of solvent accessible surface area (SASA). A few of the uncharacterized Angiogenin mutants, such as - D22G (detected in a large United States ALS-phenotype cohort) and L35P (arising through a T195C single nucleotide polymorphism), have been expressed in E.coli and purified. A perfect match of functional experiment results with MD simulation validates our method of predicting benign or deleterious ANG mutations.

Representation of human Angiogenin showing its functional sites and D22G, L35P mutations. The D22G and L35P mutations in Angiogenin (PDB ID: 1B1I) are labeled and represented as stick models (red). The three key functional sites of human Angiogenin (catalytic triad, nuclear localization signal and receptor binding site) are also shown. The color scheme is as follows - Angiogenin protein: yellow, nuclear localization signal: green, receptor-binding site: violet and catalytic triad of Angiogenin: yellow and blue stick models.


Loss-of-function predictions from MD simulations for D22G and L35P-Angiogenin mutants and their validation through ribonucleolytic and nuclear translocation activity assays. The predicted loss of ribonucleolytic activity for D22G Angiogenin mutant and loss of both ribonucleolytic and nuclear translocation activities for L35P mutant from MD simulations completely corroborated our MD simulation-based predictions, suggesting the proposed molecular mechanism(s) obtained from simulations are accountable for loss of ribonucleolytic and nuclear translocation activities.