monsoon-nlp
commited on
Commit
•
70c3731
1
Parent(s):
f1cf24c
avoid confusion on split/set
Browse files
README.md
CHANGED
@@ -22,7 +22,7 @@ Amino acid sequences use [IUPAC-IUB codes](https://en.wikipedia.org/wiki/FASTA_f
|
|
22 |
XML source from https://www.uniprot.org/help/downloads <br/>
|
23 |
CoLab notebook: https://colab.research.google.com/drive/1M6sO0Ws6i5z9VUXIXopiOqo1OkQ7K-1g?usp=sharing
|
24 |
|
25 |
-
## Pretraining
|
26 |
|
27 |
Amino acid sequences for **unreviewed proteins (TrEMBL)**
|
28 |
|
@@ -31,7 +31,7 @@ Large proteins are split into a new row every 8,000 letters.
|
|
31 |
|
32 |
### Share of taxa
|
33 |
|
34 |
-
In the pretraining
|
35 |
- 31% Papilionoideae/Faboideae (soybeans, peas, pulses, peanuts)
|
36 |
- 19% Triticeae (wheat, barley, rye, and relatives)
|
37 |
- 17% Oryzeae (rice, wild rice, and relatives)
|
@@ -47,7 +47,7 @@ In the pretraining set:
|
|
47 |
- 0.2% Beta vulgaris (beet)
|
48 |
- 0.1% Bambusa (includes edible bamboo shoots)
|
49 |
|
50 |
-
## Finetuning
|
51 |
|
52 |
**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.
|
53 |
|
@@ -61,18 +61,18 @@ Annotations may be empty, or may include information such as:
|
|
61 |
|
62 |
Removed PubMed reference numbers to avoid training models to hallucinate PubMed references.
|
63 |
|
64 |
-
In the current state of plant genomics research, about half of the
|
65 |
-
a fifth are from Papilionoideae/Faboideae, and seven taxa have less than 1
|
66 |
|
67 |
-
## Research
|
68 |
|
69 |
-
Formatted like the finetuning
|
70 |
|
71 |
|
72 |
|
73 |
## Limitations and Safety Notes
|
74 |
|
75 |
-
Proteins and review status on UniProt are from
|
76 |
|
77 |
You should at least pretrain on more than *Oryza sativa*, because its reviewed proteins are probably some of the most common/important.
|
78 |
|
|
|
22 |
XML source from https://www.uniprot.org/help/downloads <br/>
|
23 |
CoLab notebook: https://colab.research.google.com/drive/1M6sO0Ws6i5z9VUXIXopiOqo1OkQ7K-1g?usp=sharing
|
24 |
|
25 |
+
## Pretraining split
|
26 |
|
27 |
Amino acid sequences for **unreviewed proteins (TrEMBL)**
|
28 |
|
|
|
31 |
|
32 |
### Share of taxa
|
33 |
|
34 |
+
In the pretraining split:
|
35 |
- 31% Papilionoideae/Faboideae (soybeans, peas, pulses, peanuts)
|
36 |
- 19% Triticeae (wheat, barley, rye, and relatives)
|
37 |
- 17% Oryzeae (rice, wild rice, and relatives)
|
|
|
47 |
- 0.2% Beta vulgaris (beet)
|
48 |
- 0.1% Bambusa (includes edible bamboo shoots)
|
49 |
|
50 |
+
## Finetuning split
|
51 |
|
52 |
**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.
|
53 |
|
|
|
61 |
|
62 |
Removed PubMed reference numbers to avoid training models to hallucinate PubMed references.
|
63 |
|
64 |
+
In the current state of plant genomics research, about half of the finetuning split are from rice and related
|
65 |
+
species, a fifth are from Papilionoideae/Faboideae, and seven taxa have less than 1% each.
|
66 |
|
67 |
+
## Research split (TODO)
|
68 |
|
69 |
+
Formatted like the finetuning split, but with proteins from quinoa and amaranth.
|
70 |
|
71 |
|
72 |
|
73 |
## Limitations and Safety Notes
|
74 |
|
75 |
+
Proteins and review status on UniProt are from March 29, 2024.
|
76 |
|
77 |
You should at least pretrain on more than *Oryza sativa*, because its reviewed proteins are probably some of the most common/important.
|
78 |
|