Research map / Time series / Foundation models for time series
TwinWeaver: An LLM-Based Foundation Model Framework for Pan-Cancer Digital Twins
Nikita Makarov, Maria Bordukova, Lena Voith von Voithenberg, Estrella Pivel-Villanueva, Sabrina J Mielke, Jonathan Wickes, Hanchen Wang, Mingyu Derek Ma, Keunwoo Choi, Kyunghyun Cho, Stephen Ra, Raul Rodriguez-Esteban and 2 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| serialize patient history into text and fine-tune an LLM (GDT) versus time-series baselines | Llama 3.1 8B Instruct · FH-FMI CGDB blood biomarker forecasting | median MASE: 0.966 -> 0.867 | Improved |
| GDT versus RSF for risk ranking | Llama 3.1 8B Instruct · survival, progression, therapy switching | IPCW C-index: 0.662 -> 0.703 | Improved |
| GDT versus RSF on time-to-metastasis with few training indications | Llama 3.1 8B Instruct · time-to-metastasis (N = 3 indications) | IPCW C-index: 0.563 -> 0.502 | Hurt |
| fine-tune GDT on OAK/IMpower131 trials instead of zero-shot | GDT · POPLAR and IMpower130 event prediction | IPCW C-index: 0.656 -> 0.672 | Improved |
| fine-tune GDT with few trial samples per variable (ablation) | GDT · POPLAR forecasting (trained on OAK) | MASE | Mixed |
| fine-tune GDT with 64 or more samples per variable (ablation) | GDT · POPLAR forecasting (trained on OAK) | MASE | Improved |
| add reasoning chains (SFT then GRPO with negative MAE reward) (ablation) | GDT · NSCLC neutrophil forecasting | MASE: 0.828 -> 0.862 | Hurt |
| use full patient history instead of baseline information only (ablation) | GDT · NSCLC one-year survival (history of at least 52 weeks) | IPCW C-index: 9.9% improvement | Improved |
| pan-cancer training across indications (ablation) | GDT · individual cancer indications | performance | Improved |
| CLMBR-T fixed code vocabulary versus text serialization | CLMBR-T · FH-FMI CGDB event prediction | IPCW C-index | Hurt |
Closest papers by meaning
- MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology (NeurIPS 2025 · similarity 0.86)
- REACT-LLM: A Benchmark for Evaluating LLM Integration with Causal Features in Clinical Prognostic Tasks (AAAI 2026 · similarity 0.86)
- Med-Real2Sim: Non-Invasive Medical Digital Twins using Physics-Informed Self-Supervised Learning (NeurIPS 2024 · similarity 0.86)
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.84)
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · similarity 0.84)
- CLIMB: Data Foundations for Large Scale Multimodal Clinical Foundation Models (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
