File size: 4,274 Bytes
128a21a
 
ad1e504
 
 
 
 
f1cf24c
496a8c6
f1cf24c
 
 
496a8c6
 
f1cf24c
 
99d26f0
 
 
 
496a8c6
 
7cc45a1
 
128a21a
ad1e504
 
 
fe995ce
fcf7903
ad1e504
af67f30
 
 
 
 
fe995ce
ad1e504
 
70c3731
ad1e504
 
 
f1cf24c
 
ad1e504
 
 
70c3731
ad1e504
 
 
 
 
 
 
 
 
 
 
 
 
 
 
70c3731
ad1e504
 
 
 
 
 
 
 
 
 
 
 
 
70c3731
 
ad1e504
99d26f0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7cc45a1
ad1e504
7cc45a1
ad1e504
7cc45a1
 
ad1e504
 
 
70c3731
ad1e504
 
 
fcf7903
 
 
ad1e504
 
99d26f0
f1cf24c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
---
license: cc
tags:
- protein
- biology
- botany
pretty_name: GreenBeing
configs:
- config_name: pretraining
  data_files:
  - split: pretraining
    path: "proteins_pretraining.csv"
- config_name: finetuning
  data_files:
  - split: finetuning
    path: "proteins_finetuning.csv"
- config_name: evaluation
  data_files:
  - split: evaluation
    path: "proteins_evaluation.csv"
- config_name: research
  data_files:
  - split: research
    path: "proteins_research.csv"
---

# GreenBeing Proteins dataset

Proteins from UniProtKB (knowledge base), from select food crops and related species. <br/>
Amino acid sequences use [IUPAC-IUB codes](https://en.wikipedia.org/wiki/FASTA_format#Sequence_representation) where letters A-Z map to amino acids. <br/>

Usage (due to different schema on splits):
```
load_dataset("monsoon-nlp/greenbeing-proteins", "pretraining", split="pretraining")
```

XML source from https://www.uniprot.org/help/downloads <br/>
CoLab notebook: https://colab.research.google.com/drive/1M6sO0Ws6i5z9VUXIXopiOqo1OkQ7K-1g?usp=sharing

## Pretraining split

Amino acid sequences for **unreviewed proteins (TrEMBL)**

Each row contains a species or subspecies name (for filtering), and the amino acid sequence of a protein.
Large proteins are split into a new row every 8,000 letters.

### Share of taxa

In the pretraining split:
- 31% Papilionoideae/Faboideae (soybeans, peas, pulses, peanuts)
- 19% Triticeae (wheat, barley, rye, and relatives)
- 17% Oryzeae (rice, wild rice, and relatives)
- 12% Paniceae (most millets)
- 6% Solanum (potatoes, tomatoes, other nightshades, etc.)
- 5% Zea (corn/maize)
- 2% Sorghum
- 2% Lactuca sativa (lettuce)
- 2% Capsicum (chili peppers)
- 2% Cucurbita (squash, pumpkins)
- 0.8% Spinacia
- 0.7% Asparagus
- 0.2% Beta vulgaris (beet)
- 0.1% Bambusa (includes edible bamboo shoots)

## Finetuning split

**Reviewed proteins (Swiss-Prot)** from above taxa. Each row contains a gene name, species or subspecies, an amino acid sequence, and comments / annotations available in UniProt.

A gene name may match multiple entries on UniProt from different accessions.

Annotations may be empty, or may include information such as:

- likely function, written in English
- location inside of cell (e.g. "Subcellular locations: Vacuole")
- locations in the plant (e.g. "Expressed in roots, stems")

Removed PubMed reference numbers to avoid training models to hallucinate PubMed references.

In the current state of plant genomics research, about half of the finetuning split are from rice and related 
species, a fifth are from Papilionoideae/Faboideae, and seven taxa have less than 1% each.

## Evaluation split

**Reviewed proteins (Swiss-Prot)** from other genera (avocado, carrot, cassava, lychee, prunus (i.e. peaches, cherries, plums)).
Each row contains a gene name, species or subspecies, an amino acid sequence, and comments / annotations available in UniProt.

A gene name may match multiple entries on UniProt from different accessions.

Annotations may be empty, or may include information such as:

- likely function, written in English
- location inside of cell (e.g. "Subcellular locations: Vacuole")
- locations in the plant (e.g. "Expressed in roots, stems")

Removed PubMed reference numbers to avoid training models to hallucinate PubMed references.

## Research split

Proteins from quinoa (~99%), cañihua, and the three species of amaranth which are currently grown for grain.

Columns contain UniProt/TrEMBL gene names, species, and the amino acid sequence.
Large proteins are split into a new row every 8,000 letters.

## Limitations and Safety Notes

Proteins and review status on UniProt are from March 29, 2024.

You should at least pretrain on more than *Oryza sativa*, because its reviewed proteins are probably some of the most common/important.

Understanding reviewed/unreviewed proteins: https://www.uniprot.org/help/uniprotkb_sections

The reviewed sequences are often similar to unreviewed sequences from related species and accessions, and are not suitable to test predicting/infilling/completion.

Species include inedible wild relatives.<br/>
Some people have allergic reactions to wheat/gluten, nightshades, maize, lychee, and other crops.<br/>
Chili peppers can be painfully spicy.