model card: revised pipeline is screen -> gene gate -> LLM; cross-encoder off the critical path
Browse files
README.md
CHANGED
|
@@ -18,9 +18,24 @@ language:
|
|
| 18 |
Fine-tuned [`ncbi/MedCPT-Cross-Encoder`](https://huggingface.co/ncbi/MedCPT-Cross-Encoder)
|
| 19 |
that scores a PubMed **title + abstract** against a **Gene2Phenotype (G2P) LGMDE thread**
|
| 20 |
(one gene–disease entry rendered as a 15-field string) and returns a sigmoid relevance score
|
| 21 |
-
in [0, 1]. It
|
| 22 |
-
([biomedicalinformaticsgroup/LitDD_mining](https://github.com/biomedicalinformaticsgroup/LitDD_mining))
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
The previous release is preserved at tag [`v1-original`](../../tree/v1-original). It is
|
| 26 |
superseded: on the current panel rendering it calls essentially every pair positive at 0.5
|
|
|
|
| 18 |
Fine-tuned [`ncbi/MedCPT-Cross-Encoder`](https://huggingface.co/ncbi/MedCPT-Cross-Encoder)
|
| 19 |
that scores a PubMed **title + abstract** against a **Gene2Phenotype (G2P) LGMDE thread**
|
| 20 |
(one gene–disease entry rendered as a 15-field string) and returns a sigmoid relevance score
|
| 21 |
+
in [0, 1]. It was the candidate-ranking stage of the LitDD pipeline
|
| 22 |
+
([biomedicalinformaticsgroup/LitDD_mining](https://github.com/biomedicalinformaticsgroup/LitDD_mining)).
|
| 23 |
+
|
| 24 |
+
> ### ⚠️ Not required by the current pipeline (2026-09-02)
|
| 25 |
+
>
|
| 26 |
+
> The revised pipeline is **screen → gene gate → LLM adjudication**: each abstract is restricted
|
| 27 |
+
> to the Gene2Phenotype entries of the genes actually mentioned in its title+abstract (mean 2.4
|
| 28 |
+
> candidates, median 2), and **all** of those candidates are shown to the LLM with no score
|
| 29 |
+
> threshold. Measured on the held-out annotated test split, that gene gate offers every curated
|
| 30 |
+
> entry for 606/606 curated abstracts (retrieval recall 1.000), and removing the cross-encoder
|
| 31 |
+
> changes end-to-end performance by less than the noise (P 0.840 / R 0.848 without it vs
|
| 32 |
+
> P 0.851 / R 0.850 with it, n = 2,731 abstracts, McNemar p = 0.86).
|
| 33 |
+
>
|
| 34 |
+
> This model is retained because it is **required to reproduce the published pipeline**, and it
|
| 35 |
+
> remains useful as (i) an optional ranker when an abstract yields many candidates and a cap is
|
| 36 |
+
> wanted, and (ii) a confidence signal for auditing LLM answers (keeping only answers whose
|
| 37 |
+
> entries score ≥ 0.95 raises answer precision 0.927 → 0.936 while discarding 15% of answers).
|
| 38 |
+
> It is no longer part of the deployed critical path, and no score gate is applied.
|
| 39 |
|
| 40 |
The previous release is preserved at tag [`v1-original`](../../tree/v1-original). It is
|
| 41 |
superseded: on the current panel rendering it calls essentially every pair positive at 0.5
|