Internal server (500) errors for datasets with "inf" values for Beta or std_err_beta

[abought]

Summary

JSON does not support serializing values of "infinity". Trying to do so results in errors on the client, server, or both.

When a GWAS file has a value of infinity in these columns, plots of that region fail to render with a 500 Internal Server Error (see screenshot) because the backend json serializer fails.

Proposed fix

We already special-case serialization of these values for the neg_log_pvalue field. Similar treatment of beta and stderr_beta should alleviate this error. This should be tested on a real dataset, in case there are follow on problems (UI etc) when displaying plot tooltips with inf values.

locuszoom-hosted/locuszoom_plotting_service/api/serializers.py

Lines 46 to 61 in d726fb2

	beta = drf_serializers.FloatField(read_only=True)
	se = drf_serializers.FloatField(source='stderr_beta', read_only=True)
	alt_allele_freq = drf_serializers.FloatField(read_only=True)
	def get_neg_log_pvalue(self, row):
	"""
	Many GWAS programs suffer from underflow and may represent small p=0/-logp=inf
	The JSON standard can't handle "Infinity", but the string 'Infinity' can be type-coerced by JS, eg +value
	Therefore we serialize this as a special case so it can be used in the frontend
	"""
	value = row.neg_log_pvalue
	if value is not None and math.isinf(value):
	return 'Infinity'
	else:
	return value

We've seen several reports of this problem recently, after a long period of none. It's possible that a new GWAS program has been released that is prone to under/overflow, which might make this a higher priority for users in the future.