Newton-Conover Health and Science High School

This paper introduces a unique Chatbot book club application, harnessing Large Language Models (LLMs) to prevent cognitive decline in seniors, as part of a pilot study conducted by IBM and Harvard Medical School. The study will follow two groups of 70+ years old in assisted living facilities - a control arm and a group encouraged to read books. The Chatbot engages participants in the reading group in conversations about the books to promote continued engaged reading, with the hypothesis that this will measurably improve cognitive function. Utilizing prompt engineering and user profiles, it generates personalized, thought-provoking questions. The friendly user interface with large fonts and voice interactivity encourages cognitive engagement. Future work will deploy the Chatbot in the year-long study, iterate on improvements, and explore broader applications.

Human β defensin type 2 (hBD-2), a cationic cysteine-rich peptide secreted from the human innate immune system, can bind Spike-RBD at the same site as receptor ACE2, thus blocking viral entry into ACE2-expressing cells. In order to find out the impact of CoV-2 mutations on hBD-2's antiviral activity, it is important to investigate the binding and interaction of hBD-2 with RBD mutants. All-atom molecular dynamics simulations were conducted on typical RBD mutants, including N501Y, E484K, P479S, T478I, S477N, N439K, K417N, and N501Y-E484K-K417N, binding with hBD-2. Starting from the stable binding structure of hBD-2 and wt-RBD and ClusPro and HADDOCK docking-predicted initial structures, the RBD variants bound with hBD-2 simulations were set up, and NAMD simulations were conducted. Based on the structure and dynamics analysis, it was found that most RBD variants can still form a similar number of hydrogen bonds with hBD-2, in addition to having a similar-sized buried surface area (BSA) and a similar binding interface to the RBD wildtype. However, the RBD triple mutant formed a less stable binding structure with hBD-2 than other variants. Additionally, the free energy perturbation (FEP) method was applied to calculate the contribution of key mutant residues to the binding and the free energy change caused by the mutations. The result shows that N439K, K417N, and the trimutation increase the binding free energy of RBD with hBD-2; thus, RBD should bind less stably with hBD-2. E484K decreases the binding free energy, thus it should bind more stably with hBD-2, while N501Y, S477N, T478I, and P479S almost do not change the binding free energy with hBD-2. The MM-GBSA method predicted the binding interaction energy which shows that the trimutant should be able to escape the binding with hBD-2 but N501Y should not. The result can provide insight into understanding the functional mechanism of hBD-2 combating SARS-CoV-2 mutants.

The ability to remember previous tasks while concomitantly building upon prior knowledge to acquire new skills is a vital aspect of human learning. Artificial Intelligence algorithms, however, have traditionally struggled in this respect, mastering a specific task but being unable to transfer this knowledge to new tasks or forgetting previous skills when learning new ones. We examine ways in which the weights in neural networks can be constrained to ensure that the networks can acquire new skills and also remember previous tasks, allowing them to learn multiple tasks sequentially. We test our approach by building a network that learns how to play two different Atari 2600 games. Although some approaches lead to varying amounts of forgetting, others yield successful results that even beat human performance on some games.