Development of an AI-Enhanced Extended Reality Learning Environment for Biotechnology Laboratories: A Secondary-Data-Driven Design and Risk-Aware Evaluation

Abstract

Extended reality (XR) can provide repeatable, low-risk rehearsal of biotechnology laboratory procedures, yet existing virtual laboratories seldom combine adaptive artificial intelligence (AI), explicit safety control, process-aware analytics, protocol-level domain fidelity, and auditable governance. This study develops BioXR-AI, a safety-constrained XR information system, through a structured secondary evidence map and computational verification. Thirty-six peer-reviewed studies published from 2008 to 2025 were coded across ten capabilities spanning biotechnology context, immersion, virtual laboratories, adaptive AI, analytics, safety, procedural assessment, instructional accessibility, governance, and complementarity with physical laboratories. Evidence-weighted gap analysis identified safety scaffolding (weighted prevalence = 0.080), biotechnology specificity (0.125), governance and explainability (0.210), adaptive AI (0.236), and learning analytics (0.248) as high-priority design requirements. No included study combined biotechnology context, adaptive AI, immersive XR, explicit safety control, analytics, and governance in one learning environment. These gaps informed a layered architecture comprising a protocol digital twin, uncertainty-aware event interpretation, risk-sensitive feedback, grounded explanatory AI, learning analytics, accessibility support, and faculty-controlled governance. The architecture was evaluated using 7,200 synthetic protocol traces covering eight biotechnology procedures and ten deviation classes. On two held-out procedures, BioXR-AI achieved F1 = 0.951, critical-error recall = 0.904, and 0.46 unnecessary alerts per 100 benign traces, outperforming finite-state XR and linear-checklist baselines under explicitly stated detector assumptions. The contribution is a reproducible design-and-verification framework for safety-critical AI-XR laboratory learning. Human learning, usability, transfer, and real-world safety effects remain to be established through prospective trials.