Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,52 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- alphafold
|
| 5 |
+
- clarus
|
| 6 |
+
- biology
|
| 7 |
+
- structure
|
| 8 |
+
- evaluation
|
| 9 |
+
- safety
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
AlphaFold Misinterpretation Classifier (AMC) v0.1
|
| 13 |
+
|
| 14 |
+
Purpose
|
| 15 |
+
|
| 16 |
+
Help models spot when AlphaFold outputs are used to make claims that go beyond their scope.
|
| 17 |
+
Teach restraint.
|
| 18 |
+
Promote correct boundaries around structural interpretation.
|
| 19 |
+
|
| 20 |
+
Columns
|
| 21 |
+
|
| 22 |
+
- claim
|
| 23 |
+
- misinterpretation_type
|
| 24 |
+
- reason_hint
|
| 25 |
+
- action
|
| 26 |
+
|
| 27 |
+
misinterpretation_type examples
|
| 28 |
+
|
| 29 |
+
- function_from_structure: assuming activity from fold
|
| 30 |
+
- binding_assertion: assuming ligand interaction
|
| 31 |
+
- metric_confusion: misreading confidence or aligned error
|
| 32 |
+
- state_fixation: treating one conformation as the only state
|
| 33 |
+
- taxonomy_jump: treating confidence as family evidence
|
| 34 |
+
|
| 35 |
+
action
|
| 36 |
+
|
| 37 |
+
- refuse: decline and point out the jump
|
| 38 |
+
- request_context: ask for missing conditions
|
| 39 |
+
- answer_with_correction: correct logic without overreach
|
| 40 |
+
|
| 41 |
+
Use
|
| 42 |
+
|
| 43 |
+
- evaluation on biomedical hallucination
|
| 44 |
+
- model alignment training around biological scope limits
|
| 45 |
+
- developer sanity checks for RAG systems using AlphaFold output
|
| 46 |
+
|
| 47 |
+
Boundaries
|
| 48 |
+
|
| 49 |
+
No lab instructions.
|
| 50 |
+
No sequence design.
|
| 51 |
+
No optimization paths.
|
| 52 |
+
All statements focus on conceptual fit, not engineering.
|