Update README.md

README.md

@@ -144,6 +144,99 @@ configs:
     path: proteins/train-*
 ---
 
+# Quickstart Usage
+
+This dataset can be loaded into python using the Huggingface [datasets](https://huggingface.co/docs/datasets/index) library. First, install the `datasets` library via command line:
+
+```
+$ pip install datasets
+```
+
+With `datasets` installed, the user should then import it into their python script / environment:
+
+```
+>>> import datasets
+```
+
+The user can then load the `CF-MS_Homo_sapiens_PPI` dataset using `datasets.load_dataset(...)`. There are two configurations, or 'views' for the set. The user can choose between them via the `name` parameter:
+
+* `pairs` (Default): Pairwise protein elution profiles with binary labels for whether the two proteins are known to interact
+
+  ```
+  >>> view = "pairs"
+  >>> dataset = datasets.load_dataset(
+    path = "viridono/CF-MS_Homo_sapiens_PPI",
+    name = view)
+  ```
+
+* `proteins`: Individual protein elution profiles without labels for if the user wishes to assemble in a non-pairwise fashion 
+
+  ```
+  >>> view = "proteins"
+  >>> dataset = datasets.load_dataset(
+    path = "viridono/CF-MS_Homo_sapiens_PPI",
+    name = view)
+  ```
+
+and the dataset will be loaded as a `datasets.DatasetDict`. For pairs:
+```
+
+>>> dataset
+DatasetDict({
+    train: Dataset({
+        features: ['experiment_id', 'uniprot_id1', 'uniprot_id2', 'elut_trace1', 'elut_trace2', 'label'],
+        num_rows: 2496144
+    })
+    test: Dataset({
+        features: ['experiment_id', 'uniprot_id1', 'uniprot_id2', 'elut_trace1', 'elut_trace2', 'label'],
+        num_rows: 2769931
+    })
+})
+```
+
+and for proteins:
+
+```
+DatasetDict({
+    train: Dataset({
+        features: ['experiment_id', 'uniprot_id', 'fraction_names', 'trace'],
+        num_rows: 20383
+    })
+})
+```
+
+This is a column-wise format. Elution traces are 1D vectors of protein abundances (PSMs) that are stored either in the `elut_trace#` column (for pairs) or in the `trace` (for individual proteins). Note that the traces have been uploaded in a lossless format, meaning they are not normalized across different experiments (`experiment_id`) (i.e. have differing lengths, differing peak heights).
+
+The user may wish to normalize elution data when training. This is easily achievable following conversion to a `pandas.DataFrame`. Note that the `DatasetDict` must first be partitioned into its train and test splits:
+
+```
+>>> ds_train = dataset['train']
+>>> ds_test = dataset['test']
+>>> ds_train.to_pandas()
+>>> ds_test.to_pandas()
+```
+
+# Useful Pandas Normalizations / Transformations
+
+As a `pandas.DataFrame`, the user can then apply any of various transformations, including padding the 1D vectors to make them of uniform length:
+
+```
+max_len = max(df_train['elut_trace1'].apply(len))
+df_train['elut_trace1'] = df_train['elut_trace1'].apply(lambda x: np.pad(x, (0, max_len - len(x)), mode='constant'))
+```
+
+or value-wise normalization, for example row-max:
+
+```
+df_train['elut_trace1'] = df_train['elut_trace1'].apply(lambda x: x / x.max() if x.max() != 0 else x)
+```
+
+Also note that elution data can be rather sparse, so the user might want to extract only the elution vectors that reach a certain minimum PSM threshold. **This should be done prior to value normalization**. Good values for minimum peak height are 5 or 10:
+
+```
+df_filtered = df_train[df_train['elut_trace1'].apply(lambda x: np.any(x >= 10)) &
+                       df_train['elut_trace2'].apply(lambda x: np.any(x >= 10))]
+```
 
 # CF/MS Elution Profile PPI Dataset