flair-bio/meltome_human

Meltome human: apparent melting temperature

A standardized human Meltome Atlas regression dataset, derived from FLIP at commit 62cace8. The target is experimentally measured apparent melting temperature in degrees Celsius, not organism growth temperature or a binary stability label.

Data

Split	Measurements
Train	7,078
Validation	844
Test	1,886

Column	Description
id	Unique measurement identifier.
sequence	Amino-acid sequence, without truncation.
targets	Apparent melting temperature in degrees Celsius (floating point).
split	Standardized train, validation, or test assignment.
set	Upstream train/test designation.
source_validation	Upstream validation indicator.
source_split	Original train/validation/test assignment.
source_row	Source row index.

Sequences contain only the 20 canonical amino acids and range from 39 to 2,047 residues. Noncanonical and overlength records were excluded, not cropped. Released human train, validation, and test assignments are preserved. Repeated experimental measurements of the same sequence within a split are preserved. There are no exact sequence overlaps between splits; this alone does not establish homology separation.

Evaluation and limitations

Use sequence regression and report pooled Spearman correlation across test measurements. Select hyperparameters on validation, never test. Repeated measurements remain separate evaluation rows.

Apparent melting temperature depends on experimental context; it is not a context-independent measure of intrinsic stability. Treat human and mixed as separate benchmarks rather than independent datasets for combined scoring.

Usage

from datasets import load_dataset

dataset = load_dataset("flair-bio/meltome_human")

Consult the upstream source for licensing and attribution requirements.