Hands-on training for research groups

This material has been taught in university courses and small groups. As an RNA Forecast training it is being prepared, and dates will be announced to the waitlist first.

Who it is for. Experimental RNA labs that want computational depth; biotech teams working on mRNA, siRNA or structured RNA; doctoral programmes; tool developers.

What is covered, chosen per group, half day to full day:

• RNA secondary structure and the nearest-neighbour model: what RNAfold computes, and why the MFE structure is one answer among many.
• The Boltzmann ensemble: pair probabilities, centroid and MEA structures, ensemble diversity; reading a dot plot; when the MFE misleads.
• Probing data: SHAPE and DMS reactivities as soft constraints, the three ViennaRNA methods, what the data change and what they do not.
• Folding kinetics: suboptimal structures, barrier trees, barriers and treekin, cotranscriptional folding; when kinetics matter for a design.
• Accessibility and RNA–RNA interaction: RNAplfold, RNAcofold and RNAup for siRNA, ASO and miRNA sites.
• Design review before synthesis: objective, ensemble defect, kinetic traps, variant robustness; a checklist participants take home.

Format. Live, remote or on site, groups of 6 to 12, half day or full day, participants' own sequences welcome. Materials stay with the group.

Free material on the topic: interpreting SHAPE and chemical probing data and the RNA structure glossary.

New. The setup guides this course is built on are published and maintained on michaelwolfinger.com; the course itself is in preparation and will run as a live half-day cohort, updated for every run because the tools change monthly.

A scientist's workshop on running coding agents, Claude Code and OpenAI Codex CLI, against models that run on the group's own hardware with LM Studio or Ollama, on Apple Silicon and on Linux with NVIDIA or AMD GPUs, with notes for Windows. For groups whose unpublished data, manuscripts and code must not leave the machine.

What is covered:

• Hardware and model choice: what fits in 16, 24, 32 and 64+ GB; measured memory needs; tool calling as the property that decides whether a model is usable.
• Setup on macOS and Linux, and what differs on Windows: LM Studio, Ollama, the environment variables, context length as the setting that breaks most sessions.
• A research workflow: scripts and analysis notebooks, tests as the safety net, AGENTS.md and CLAUDE.md for durable instructions, reading diffs.
• What local models can and cannot do compared with hosted ones; where they fail and how to work around it.
• Privacy and cost: what leaves the machine with each configuration, telemetry switches, what a GPU box costs against API spend.
• Hands-on: every participant leaves with a working setup on their laptop or against the group's server.

Prerequisites. A laptop with 16 GB or more, or access to a group machine; comfort with a terminal. Format. Live cohort, half day, 6 to 12 participants, remote or on site.

The guides: Claude Code with a local model on Apple Silicon, Claude Code with Ollama, Codex CLI with LM Studio or Ollama.