Update README.md

README.md

@@ -1,3 +1,37 @@
----
-license: mit
----
+---
+license: mit
+
+---
+
+# Description
+
+Thermostability prediction is a regression task where each input protein *x* is mapped to a label *y* ∈ *R*, corresponding to the thermostability of *x*.
+
+# Splits
+
+**Structure type:** AF2
+
+The dataset is from [**FLIP: Benchmark tasks in fitness landscape inferencefor proteins**](https://www.biorxiv.org/content/10.1101/2021.11.09.467890v2). We employ all proteins from the“Human-cell”splits (proteins that lack AF2 structures are removed), and split them based on 70% structure similarity (see [ProteinShake](https://github.com/BorgwardtLab/proteinshake/tree/main)), with the number of training, validation and test set shown below:
+
+- Train: 5310
+- Valid: 706
+- Test:  706
+
+# Data format
+
+We organize all data in LMDB format. The architecture of the databse is like:
+
+**length:** The number of samples
+
+**0:** 
+
+- **name:** The UniProt ID of the protein
+
+- **seq:** The structure-aware sequence
+- **plddt**: pLDDT values at all positions
+- **fitness:** fitness label of the sequence
+
+**1:**
+
+**···**
+