Fine-tuned model from the PsyNamic project (code, webapp): Study Control.
This is the model deployed in the PsyNamic webapp after the revision (Nov 2026). It supersedes STRIDE-lab/pubmedbert-study-control-20250221 (tag submission-v1, initial submission). All current models: PsyNamic collection.
microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract
Study Control (multi-label: several labels per abstract possible)
Describes the type of control used in the study (e.g., placebo, active control, non-controlled).
| ID | Label | Definition |
|---|---|---|
| 0 | Placebo | – |
| 1 | Active control | E.g., midazolam as a control for ketamine. |
| 2 | Pre-post | A study design where measurements are taken on the same subjects at two points in time: before (pre) and after (post) a psychedelic intervention. |
| 3 | Other control | E.g., non-users, cross-over. |
| 4 | Non-controlled | – |
| 5 | Not applicable | E.g., for systematic reviews. |
Multi-label: a sigmoid is applied per label, and every label with probability ≥ 0.29 is predicted. This is the threshold used in the PsyNamic webapp. It was tuned on the validation set with a pre-specified rule: start at the threshold with maximal weighted F1, then lower it as long as weighted precision stays within 0.05 of its value there.
Five base models (BiomedBERT-abstract, BioBERT, BioLinkBERT, ClinicalBERT, SciBERT) were fine-tuned with identical hyperparameters. The checkpoint is the one with the lowest validation loss (early stopping), and the base model was selected on the validation set (weighted F1). The test set was used once, after selection.
Held-out test set (98 abstracts), tuned threshold as in the webapp, 95% bootstrap CIs (1,000 BCa resamples).
| Metric (test) | Value [95% CI] |
|---|---|
| Weighted F1 | 0.729 [0.643–0.802] |
| Weighted precision | 0.713 [0.612–0.782] |
| Weighted recall | 0.750 [0.665–0.828] |
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
repo = "STRIDE-lab/biomedbert-abstract-study-control-20261106"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSequenceClassification.from_pretrained(repo)
inputs = tokenizer("<title>. <abstract>", truncation=True, max_length=512, return_tensors="pt")
logits = model(**inputs).logits
probs = torch.sigmoid(logits)[0]
pred = [model.config.id2label[i] for i, p in enumerate(probs) if p >= 0.29]
{
"data": "<private_dataset>",
"task": "Study Control",
"model": "biomedbert-abstract",
"cross_val": false,
"batch_size": 8,
"learning_rate": 5e-05,
"weight_decay": 0.01,
"lr_scheduler": "linear",
"warmup_ratio": 0.1,
"epochs": 30,
"dropout": 0.1,
"early_stopping_patience": 5,
"gradient_clipping": 1.0,
"max_length": 512,
"threshold": 0.5,
"is_multilabel": true
}
threshold (0.5) is only the cut-off used to compute the validation metrics during training. It does not affect the model weights or the checkpoint choice (lowest validation loss). Predictions in the webapp and the test performance above use the tuned threshold 0.29 (see Prediction rule).
microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract