jwkova commited on
Commit
33fb225
·
verified ·
1 Parent(s): aabae3c

Add MIMIC dataset

Browse files
Files changed (3) hide show
  1. README.md +69 -0
  2. raw/examples.jsonl +0 -0
  3. tokenized/examples.jsonl +0 -0
README.md ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - feature-extraction
5
+ tags:
6
+ - biology
7
+ - genomics
8
+ - proteins
9
+ - multimodal
10
+ - mimic
11
+ pretty_name: LORE Examples
12
+ ---
13
+
14
+ # LORE Examples
15
+
16
+ A small set of **matched multimodal examples** from LORE, for the
17
+ [MIMIC](https://github.com/PolymathicAI/MIMIC) model — enough to try inference,
18
+ embedding, and generation across DNA, RNA, and protein modalities without wiring
19
+ up your own data.
20
+
21
+ Each example is a single biological entity (a transcript and/or its protein) with
22
+ several co-observed modalities. Rows are drawn from the held-out (validation) split
23
+ of MIMIC's training data, so they are in-distribution and length-bounded to the
24
+ model's context window.
25
+
26
+ ## Contents
27
+
28
+ Two parallel views of the **same** rows:
29
+
30
+ | Path | Format | Use |
31
+ |------|--------|-----|
32
+ | `raw/examples.jsonl` | modality name → raw value (nucleotide/AA string, per-position track, …) | `model.input([row])` — the tokenizers run for you |
33
+ | `tokenized/examples.jsonl` | `tok_<modality>` → token-id list | `model.input([row])` — pre-tokenized; needed for `prot_struct` / text modalities on a base install |
34
+
35
+ Every row carries `kind` (`rna` / `protein` / `both`) plus `uniprot_id` /
36
+ `genome_feature_id` anchors. The `tokenized` view covers **all** modalities
37
+ (including `prot_struct` and the free-text tracks); the `raw` view covers the
38
+ modalities that invert without the optional `[structure]` / `[text]` extras.
39
+
40
+ ## Usage
41
+
42
+ ```python
43
+ import json
44
+ from mimic import load_pretrained
45
+
46
+ model = load_pretrained(version="1.0")
47
+
48
+ # raw view — tokenizers run inside input()
49
+ row = json.loads(open("raw/examples.jsonl").readline())
50
+ row = {k: v for k, v in row.items() if k not in ("kind", "uniprot_id", "genome_feature_id")}
51
+ model.input([row])
52
+ reps = model.embed(sep_encodings=False)
53
+
54
+ # tokenized view — feed pre-tokenized ids (tok_ keys)
55
+ trow = json.loads(open("tokenized/examples.jsonl").readline())
56
+ trow = {k: v for k, v in trow.items() if k not in ("kind", "uniprot_id", "genome_feature_id")}
57
+ model.input([trow])
58
+ ```
59
+
60
+ ## See also
61
+
62
+ - **Model:** [`polymathic-ai/MIMIC`](https://huggingface.co/polymathic-ai/MIMIC) — the model these examples are for.
63
+ - **Code:** [`PolymathicAI/MIMIC`](https://github.com/PolymathicAI/MIMIC) (`pip install mimic-cd`).
64
+
65
+ ## Provenance & license
66
+
67
+ Derived from the validation split of MIMIC's training corpus (LORE). Released under
68
+ the MIT license, matching the model code. See the
69
+ [MIMIC repository](https://github.com/PolymathicAI/MIMIC) for details and citation.
raw/examples.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
tokenized/examples.jsonl ADDED
The diff for this file is too large to render. See raw diff