Very useful and nice changes :) I left minor comments, please ask my review again once you're done, we're almost there! Also in general, please leave a space before = symbol and after it, and also a space after punctuation like , and :

We'll look at the PR together once you implement these changes and we'll finalize the following:

• In test_standardize_graphdataset, we need to test hb_donors, for which standardize is False. And in general all the rest of the features not indicated in the dict. In order to test it, we can use _cal_mean_std with hb_donors feature; in general, we need to verify that features of dataset which are in features_transform with standardize True have mean and dev as indicated (mean around 0 and dev around 1) and different mean and dev from the same features not touched (maybe using _cal_mean_std); we also need to verify that features of dataset which are in features_transform with standardize False or which are not in the dict have mean and dev equal to the ones not touched, again maybe using _cal_mean_std
• test_feature_transform_mean_std partially tests the transformation, so we need to implement a smart way to really test transformations

I think not all the features with standardized True would got mean around 0 and dev around 1 even after transformation. For example for feature sasa, before transformation, its mean=45 & dev=41.5, and after transformation its mean=5.7 & dev=3.5.
Do you have any suggestions on setting up the mean and dev range? For now I do something like this:

for key, values in features_dict.items():
            if key in features_transform: 
                transform = features_transform.get(key, {}).get('transform')
                means = []
                devs = []
                (mean, dev) = _cal_mean_std(hdf5_path, features_transform, key)
                means.append(mean)
                devs.append(dev)
                means.append(values.mean())
                devs.append(values.std())
                if transform: #test transformed features
                    assert means[0] != means[1]
                    assert devs[0] != devs[1] 
                    assert -10 < means[0] < 10
                    assert -5 < devs[0] < 5
                else: #test hb_doners, no transformation so mean & std remain the same.
                    assert means[0] == means[1]
                    assert devs[0] == devs[1]

Very useful and nice changes :) I left minor comments, please ask my review again once you're done, we're almost there! Also in general, please leave a space before = symbol and after it, and also a space after punctuation like , and :
We'll look at the PR together once you implement these changes and we'll finalize the following:

• In test_standardize_graphdataset, we need to test hb_donors, for which standardize is False. And in general all the rest of the features not indicated in the dict. In order to test it, we can use _cal_mean_std with hb_donors feature; in general, we need to verify that features of dataset which are in features_transform with standardize True have mean and dev as indicated (mean around 0 and dev around 1) and different mean and dev from the same features not touched (maybe using _cal_mean_std); we also need to verify that features of dataset which are in features_transform with standardize False or which are not in the dict have mean and dev equal to the ones not touched, again maybe using _cal_mean_std
• test_feature_transform_mean_std partially tests the transformation, so we need to implement a smart way to really test transformations

I think not all the features with standardized True would got mean around 0 and dev around 1 even after transformation. For example for feature sasa, before transformation, its mean=45 & dev=41.5, and after transformation its mean=5.7 & dev=3.5. Do you have any suggestions on setting up the mean and dev range? For now I do something like this:

for key, values in features_dict.items():
            if key in features_transform: 
                transform = features_transform.get(key, {}).get('transform')
                means = []
                devs = []
                (mean, dev) = _cal_mean_std(hdf5_path, features_transform, key)
                means.append(mean)
                devs.append(dev)
                means.append(values.mean())
                devs.append(values.std())
                if transform: #test transformed features
                    assert means[0] != means[1]
                    assert devs[0] != devs[1] 
                    assert -10 < means[0] < 10
                    assert -5 < devs[0] < 5
                else: #test hb_doners, no transformation so mean & std remain the same.
                    assert means[0] == means[1]
                    assert devs[0] == devs[1]

Another way to verify that you're actually standardizing, is to do the inverse calculation (destandardizing the values) and verifying that the mean and the std dev of these back-transformed values are the same as before.

So in terms of code it would be, values being the ones after transformation (if present) and standardization: values_no_std = values * dev + mean
Then you can test that values_no_std.mean() and values_no_std.std() are the same as the ones obtained with the feature not standardized.