Demonstration only. Corollary is a non-commercial research demo — no warranty, no support, not for commercial use or clinical decisions. Read the disclaimer

Corollary Labs
All datasets

MSA / Training Corpus · LiteFold

Evolutionary

Precomputed MMseqs2-style evolutionary multiple sequence alignment data for protein language models and folding.

Original source

What Evolutionary is

Evolutionary is a precomputed dataset of evolutionary sequence-alignment data published by LiteFold. It is packaged to serve as drop-in multiple sequence alignment (MSA) features for protein language model training and protein structure prediction pipelines, eliminating the need to construct alignments from scratch using raw sequence databases.

Packaged into an uncompressed archive format paired with a metadata index, it allows computational biologists to query file locations and extract individual cached alignments or restore the full dataset directory tree.

How it was made

The collection consists of 134,906 indexed files comprising an original payload of 19.49 GiB. The bulk of the dataset resides in an msa_cache directory containing 134,898 compressed NumPy (.npz) files representing precomputed alignments. Additional components include an mmseqs30 directory containing FASTA files, alongside root-level files in Parquet, JSON, TSV, Bash script, and Markdown formats.

To make distribution practical, the payload is stored within a single uncompressed tar shard (shard-00000.tar) totalling 19.72 GiB. A companion metadata.csv index logs the original relative path, storage type, shard location, member path, file extension, and byte size for every file, enabling direct extraction of specific alignment members without decompressing an entire compressed archive.

Key facts

Total indexed files
134,906
MSA cache files
134,898 (.npz)
Total payload size
19.49 GiB
Archive size
19.72 GiB
Tar shards
1
Primary file formats
.npz, .fasta, .parquet, .json, .tsv, .csv

What you can use it for

Protein Structure Prediction Input
Supply precomputed evolutionary sequence alignments directly to structure-prediction networks without running iterative database searches.
Protein Language Model Training
Use cached MSA features as training data or representation inputs for evolutionary and protein language models.
Targeted Sequence Feature Extraction
Query metadata.csv to identify and extract individual .npz cache files for specific protein targets using tar member extraction.

Explore the data

No licence declared

The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.

How the columns are distributed

Over 134,906 rows of files/train

directory

string_label
  • msa_cache100%
  • mmseqs300%

0.00% empty

extension

string_label
  • .npz100%
  • .fasta0%
  • .tsv0%

filename

string_text
min
6
median
9
mean
9.29
max
22

member_path

string_text
min
11
median
19
mean
19.29
max
31

modified_utc

datetime

part_paths

float

No distribution reported for this column.

100.0% empty

parts_count

int
min
0
median
0
mean
0
max
0

path

string_text
min
11
median
19
mean
19.29
max
31

shard_path

string_label
  • shards/shard-00000.tar100%

size_bytes

int
min
543
median
144k
mean
155k
max
39M

size_human

string_text
min
5
median
10
mean
9.76
max
10

storage_type

string_label
  • tar100%

top_level

string_label
  • msa_cache100%
  • .0%
  • mmseqs300%

Strengths and limits

Good at

  • Provides ready-to-use evolutionary sequence alignments without the computational overhead of querying large sequence databases.
  • Uncompressed tar packaging allows extraction of individual MSA files directly from the archive shard.
  • Includes a comprehensive metadata.csv index enabling filtering by file extension, directory, and exact file paths.

Watch out for

  • Extracting a single file requires downloading the entire 19.72 GiB tar shard.
  • Coverage is restricted to the pre-indexed set of 134,898 cached protein alignments and does not cover arbitrary novel sequences.

Questions

Can I look inside Evolutionary without downloading it?

Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.

What licence is Evolutionary under?

Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.

What format are the multiple sequence alignments stored in?

The alignments are precomputed in an MMseqs2-style pipeline and primarily stored as 134,898 individual NumPy array (.npz) files in the msa_cache directory, supplemented by three FASTA files.

How are individual files accessed without extracting the entire archive?

Researchers can inspect metadata.csv to find the target file's member_path, download the tar shard, and open it with Python's tarfile module to extract only that specific file.

Are the tar shards compressed?

No, the archive tar shard is uncompressed, which facilitates direct streaming and extraction of specific constituent files.

Written from Evolutionary's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.

Related datasets