Reviving Professorware: An Agentic Approach

Maintaining legacy bioinformatics software remains a longstanding challenge in research computing. Many tools are developed by PhD students and remain in active use long after their graduation, even when the underlying technology stack has become outdated or unsupportable. Yet full rewrites demand substantial developer time, which research group leaders are seldom able to allocate and new recruits are rarely motivated to undertake. Large language model (LLM)-based coding assistants offer a potential new approach to this problem, but their effectiveness in real-world research software engineering (RSE) contexts remains poorly characterized.We present a case study in AI-assisted legacy modernisation using KEGGanim, a Perl CGI web application that overlays time-series gene expression data onto KEGG pathway diagrams as animated visualisations. The tool, originally developed by the BIIT group, was first published in 2008 and has remained available since then. It is currently maintained by the ELIXIR Estonia Node at the University of Tartu as one of ELIXIR Estonia’s Node Services. The tool was rewritten from scratch using Claude Code (Anthropic) as the primary development tool. The new implementation uses FastAPI, React 18, PostgreSQL, and Docker Compose, replacing static GIF output with interactive SVG animations while preserving full data pipeline logic including KEGG gene mapping, ENSG identifier conversion, and KNN imputation.The rewrite was conducted as an instrumented RSE experiment. Development sessions were logged manually for time attribution; all LLM conversation logs were preserved in JSONL format for post-hoc extraction of human intervention counts, token usage, and effort distribution across project phases. Functional correctness was validated using a four-level regression suite against reference fixtures captured from the legacy production instance.We report on the feasibility, efficiency, and failure modes of LLM-driven legacy modernisation, and discuss implications for sustainable tool development within ELIXIR and the wider RSE community.