Short Abstract: In the brain, nerve cells (neurons) communicate via a chemical messenger glutamate. Abnormalities in glutamate-dependent neural activity are the cause for many neurological diseases, including Alzheimer’s disease and stroke/brain trauma-induced nerve cell death. Glutamate has traditionally been considered as a simple, uniform molecule, existing at all brain regions and at all times as the same structure and function. We intend to establish that glutamate can exist in multiple different structure forms and thus with different levels of activity. This will not only provide new insights to our understanding of how brain functions, but may also open new doors for the design of novel strategies in the prevention and therapeutical management of neurological abnormalities, neurodegenerative diseases and neuropsychological disorders. In this study, we proposed tree-structured covalent-bond-driven molecular memetic algorithm (TCM-MA), employed for the discovery of glutamate stereoisomers. The energies as well as first and second derivatives of glutamate structure, used in this algorithm during the evolution process, are calculated using an ab-initio approach, namely, Hartree–Fock method, with the STO-3G basis set. Our computational studies discovered 524 glutamate stereoisomers. Receptor binding analysis has revealed that the binding of these glutamate stereoisomers varies greatly between 3.5 and 7.5 KCAL/MOL, indicating significant difference in the activation of the glutamate receptors. These findings encourage experimentalists to design experimental approaches that can measure and confirm the binding affinity of the stereoisomers, add a new dimension of the complexity to glutamatergic system and put forward the glutamate stereoisomers as a new field for neuroscientists to discover.

Short Abstract: The transcriptome of the brain changes during development and across brain regions, reflecting processes that determine the structure and function of its circuitry. Despite the importance of these changes, little is still known about how brain regions become specialized in terms of their transcription profiles during development. We use gene expression measures from in situ hybridization across the full developing mouse brain to quantify the specialization of regional gene expression profiles. Expression specialization is quantified as the dissimilarity in expression profiles across brain regions. Surprisingly, during the time that the brain becomes anatomically regionalized in early development, its transcription specialization decreases, reaching a low point around birth, and then rises postnatally. This hourglass shaped profile spans many genes and brain regions. The early decrease of specialization is mainly due to biological processes that are involved in constructing brain circuitry, like axon guidance, while the rising post natal specialization is largely attributed to plasticity and neural activity processes. Post natal specialization is particularly significant in the cerebellum, whose expression signature becomes increasingly different from other brain regions. This effect is also observed in the human cerebellum during the parallel developmental period, suggesting that similar specialization profiles of brain development may be abundant in mammals.

Short Abstract: The identification of nucleotide sequence variations in viral pathogens linked to disease susceptibility and clinical outcomes is important for developing vaccines and therapies, but identifying these genetic variations in rapidly evolving pathogens adapting to immune and drug selection pressures unique to each host presents a number of challenges. Machine learning tools provide new opportunities to address these challenges. In HIV infection, virus replicating within the brain causes HIV-associated dementia (HAD) and milder forms of neurocognitive impairment in 20-30% of patients with unsuppressed viremia. HIV neurotropism is primarily determined by the viral envelope (env) gene. To identify amino acid signatures in the HIV env gene predictive of HAD diagnosis, we developed a machine learning pipeline using the PART rule-learning algorithm and C4.5 decision tree inducer to train a classifier on a meta-dataset (n=860 env sequences from 78 patients: 40 HAD, 38 non-HAD). To increase the flexibility and biological relevance of our analysis, we included 4 previously published numeric factors describing amino acid hydrophobicity, polarity, bulkiness and charge, in addition to amino acid identities. Our classifier had 75% predictive accuracy in leave-one-out cross-validation. Individually, 13 (6 HAD, 7 non-HAD) of 18 amino acid signatures comprising the classifier had p-values <0.05 (Fisher’s exact test). Two HAD signatures were validated against virus from CSF of an independent cohort of 36 patients (24 HAD, 12 non-HAD). This analysis provides insight into viral genetic determinants associated with HAD, and develops novel methods for applying machine learning tools to analyze the genetics of rapidly evolving pathogens.

Short Abstract: Many biological processes can be viewed as algorithms that nature has designed to solve real-world problems. Like computational systems, these processes seek solutions that optimize similar criteria (robustness, efficiency, adaptability) and do so under similar operating principles (distributedly, often with network-based interactions, and under some cost model), which suggests a potential basis for their shared analysis.

We present our recent work on connecting a neuro-developmental process called "synaptic pruning" to computational problems in network design. During neural network development, synapses are exuberantly produced (peaking at around age 2 in humans), and then a majority of these connections (50-60%) are subsequently pruned by adolescence. This seemingly wasteful process is believed to occur to provide the brain with an opportunity to explore possible connection topologies and then fine-tune itself to ultimately produce a network that is proficient at processing information from an environment with specific demands and constraints. We mathematically formalize this problem and present results showing the benefit of pruning-based algorithms compared to other strategies for creating robust, efficient, and adaptive networks. We also discuss our efforts to experimentally quantify the rate of synapse elimination in the mouse cortex and the downstream implications of the pruning rate on the quality of the networks built. Our algorithms have many applications to the design of communication, wireless, and synthetic networks.

With recent technological advances enabling us to more deeply probe molecular and cellular systems, we believe this direction of coupling biological and computational studies to greatly expand in the future.

Short Abstract: We exploited systems biology to systematically analyze multiple high-throughput genomic studies of traumatic brain injury (TBI) to gain insights into mechanisms of action and to infer biomarker candidates.
We integrated four gene expression datasets representing distinct animal models of TBI with over 140 canonical pathways and a human protein-protein interaction network consisting of over 11,000 proteins and 80,000 interactions. We developed new computational methods to identify statistically significant pathways and protein interaction modules associated with the TBI conditions represented in the expression data. We then analyzed the TBI-specific pathway and protein interaction modules to hypothesize novel biomarkers and mechanisms of action. We experimentally validated the hypothesized biomarker through Western blot analyses using a penetrating ballistic-like brain injury (PBBI) model.

We found the statistically significant pathways (14) and protein interaction modules (27) that were up-regulated to be associated with immune response. In stark contrast, the down-regulated pathways (7) and interaction modules (6) were linked to neurological (synaptic) functions. From these, we hypothesized three protein TBI biomarkers: PSD95, DISC1, and NOS1. We confirmed, that each of these proteins was significantly (p<0.05) down-regulated through Western blot analysis of 12 in vivo experiments using a rat PBBI model (69%, 47%, and 50%, respectively). This result suggests that systems biology provides an efficient, high-yield approach to generate hypotheses that can be experimentally validated to identify novel TBI biomarkers.