Green Valley High School

Medical multiple-choice question (MCQ) answering requires specialized domain knowledge, multi-step clinical reasoning, and ability to discriminate among closely related distractor options. Current large language models (LLMs) frequently exhibit hallucinations, inconsistent reasoning patterns, and unstable performance when confronted with complex biomedical scenarios. This study introduces a structured multiagent framework integrating Retrieval-Augmented Generation (RAG) with Model Context Protocol (MCP) to enhance accuracy and reliability in medical reasoning tasks. The proposed architecture employs an administrator agent that performs domain-specific task routing and coordinates five specialized medical agents-Physiology, Pathology, Clinical Medicine, Treatment and Pharmacology, and Public Health-integrating their outputs through weighted consensus for final answer selection. Critically, this study demonstrates that knowledge database quality significantly outweighs quantity in RAG systems: curating high-quality question subsets from knowledge data yields substantially greater performance gains than utilizing entire unfiltered knowledge datasets. Evaluated on MedMCQA and MedQA-USMLE benchmarks using GPT-oss 21B and LLaMA 4Scout 17B base models without fine-tuning, the MCP-based multiagent framework achieves approximately 5% accuracy improvement (71-75%) over single-agent baselines (66-70%). Furthermore, employing only top-500 highest-quality curated questions in RAG knowledge database boosts accuracy to nearly 80%, representing $\boldsymbol{\sim} \mathbf{1 0 \%}$ overall improvement. Ablation studies confirm multi-agent consensus mechanisms provide 1.0-2.4% additional gains beyond retrieval augmentation alone, while scaling analysis reveals optimal performance at 11-13% curated knowledge data. These findings establish that domain specialization, structured agent collaboration, and strategic data curation collectively enhance both accuracy and interpretability in medical reasoning systems, offering practical pathway for deploying reliable clinical decision support without computationally expensive model fine-tuning.

Bcl-2, an anti-apoptotic protein, is involved in cancer development and progression and has been considered as a promising target in cancer therapy. ABT-199, a Bcl-2 inhibitor, has been approved, as a single agent or in combinations, for the treatment of adult patients with chronic lymphocytic leukemia (CLL), small lymphocytic lymphoma (SLL), or newly diagnosed acute myeloid leukemia (AML) patients with age of 75 years or older. Although most patients can initially benefit from ABT-199 treatment, patients eventually develop resistance majorly due to BCL2 mutations, such as G101V and D103Y, or overexpression of Bcl-xL. Therefore, inhibitors that can target both wild-type and mutant Bcl-2 as well as Bcl-xL have the potential to overcome the resistance mechanisms to ABT-199. LP-118 is a novel inhibitor targeting wild-type Bcl-2, Bcl-2G101V and Bcl-2D103Y mutants, and has moderate activity against Bcl-xL. In preclinical and clinical studies, LP-118 had excellent anticancer activity in animal models and patients with CLL or SLL. However, there is no co-crystal structure information available to show exactly how LP-118 binds to Bcl-2, Bcl-2G101V, Bcl-2D103Y, or Bcl-xL. Molecular docking is one of the most frequently used methods in drug design and discovery, due to its ability to predict the binding-conformation of small molecule ligands to the appropriate target binding site. Using the molecular docking Software ICM-Pro, we are the first to illustrate how LP-118 binds to Bcl-2, Bcl-2G101V, Bcl-2D103Y, and Bcl-xL proteins in 3D. The molecule LP-118 is tightly tugged in the reception pocket of each protein where the interactions, including hydrogen bonding, electrostatic and π-stacking interactions, slightly vary in each case. These knowledge enables a better understanding of the mechanism of actions of LP-118 and provides a foundation for the rational design of future generations of Bcl-2/Bcl-xL inhibitor that will overcome the Bcl-2 mutation resistance. Based on the docking result, we designed over 1,000 analogues of LP-118 by fine-tuning the backbone and substituents on the molecule, used ICM-Pro/VLS software to virtually screen these analogues against Bcl-2, Bcl-2G101V, Bcl-2D103Y, or Bcl-xL, and successfully identified and prioritized 10 top-ranked analogues for chemical synthesis and anticancer activity testing. Joongkyum Shin, Yao Chen, Brianna Feldmeier, Elinor Lee, Qinliang Zhao. Molecular docking and drug virtual screening of novel small molecule inhibitors of anti-apoptotic proteins for cancer treatment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4459.

e24070 Background: Patients diagnosed with cancer as adolescents or young adults (AYA: 15-39yo) have unique developmental needs (psychosocial, fertility preservation), and historically did not experience the improvements in survival observed in other age groups. Thus, the National Cancer Institute views AYA as a vulnerable group. Youth leadership students at Vestavia Hills High School (VHHS) promote philanthropy and community outreach via a service learning effort (“RISE”). RISE has supported the O’Neal Comprehensive Cancer Center AYA Oncology & Oncofertility Program since 2019. This project aimed to evaluate AYA cancer knowledge among area high school students and the association between RISE participation and AYA cancer awareness. Methods: A student designed cross-sectional electronic survey was distributed by student champions to current students or recent graduates from Birmingham-area (AL) high schools. Surveys captured demographics, AYA cancer knowledge, information sources, attitudes, and perceptions. VHHS students were also asked about RISE participation. Multivariable logistic regression examined the association between RISE participation and AYA cancer knowledge. Results: The majority of survey respondents (n = 87), were in 12 th grade (71.3%) with 14.9% recent graduates. The majority were 17yo (51.7%) and female (56.3%). Most were white (88.5%) and non-Hispanic (96.5%). VHHS represented 54% of students, among whom 95.7% were RISE participants. Most students accurately defined cancer (95.4%), considering it extremely serious (87.4%). While most (94.3%) had heard of cancer in AYAs, less (47.1%) believed causes differ from those in adults > 40. Over half (51.7%) were aware of early signs and symptoms and considered it curable (55.2%). The most common information sources were school (67.8%), internet/social media (73.6%), and family (79.3%), with 78.2% considering these sources trustworthy. Many believed schools should teach more about cancer (88.5%) and were interested in learning more (81.6%). RISE participation was associated with higher odds of recognising AYAs as a unique cancer group (adjusted odds ratio = 2.8, 95% Confidence Interval: 1.1-7.4; p = 0.04) adjusting for age, sex, and race. Conclusions: Adolescents are aware of the prevalence and potential signs of cancer in AYAs. However, high school students exposed to an AYA cancer campaign were more likely to recognise AYAs as a unique cancer group. Although schools were identified as a key source of information, the majority of students believed further efforts by schools could increase AYA cancer knowledge and awareness and expressed their willingness to learn more. This work underscores schools as an effective platform for raising AYA cancer awareness among high school-aged students.