File size: 5,603 Bytes
e98e7ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7f1fb27
e98e7ab
 
 
7f1fb27
e98e7ab
 
 
7f1fb27
e98e7ab
 
 
7f1fb27
e98e7ab
 
 
 
4f59ed1
e98e7ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2f4d415
e98e7ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
---
license: cc-by-4.0
task_categories:
  - question-answering
language:
  - en
tags:
  - chemistry
  - solubility
  - cheminformatics
  - llm-benchmark
  - smiles
pretty_name: SoluBench
configs:
  - config_name: task1
    data_files:
      - split: test
        path: task1_pairwise_solvent_comparison.csv
  - config_name: task2
    data_files:
      - split: test
        path: task2_best_solvent_selection.csv
  - config_name: task3
    data_files:
      - split: test
        path: task3_cosolvent_effect_prediction.csv
  - config_name: task4
    data_files:
      - split: test
        path: task4_pairwise_compound_comparison.csv
---

# SoluBench

**SoluBench** is a benchmark for evaluating large language models on solubility-related tasks of various complexity. It is built on top of [BigSolDB v2.0](https://www.nature.com/articles/s41597-025-05559-8) and [MixtureSolDB](https://doi.org/10.26434/chemrxiv-2025-m51v8) — two curated experimental solubility datasets.

> 📄 Preprint: *coming soon*  
> 💻 GitHub: [levakrasnovs/SoluBench](https://github.com/levakrasnovs/SoluBench)

---

## Tasks

| Config | Task | Description | Input | Output | n | Random baseline |
|--------|------|-------------|-------|--------|---|-----------------|
| `task1` | Pairwise solvent comparison | Given a compound (SMILES) and two solvents, select the one with higher solubility | SMILES + 2 solvents | A / B | 3699 | 50.0% |
| `task2` | Best solvent selection | Given a compound and a list of solvents, select the one with highest solubility | SMILES + N solvents | letter | — | ~17.1% |
| `task3` | Co-solvent effect prediction | Given a compound, a base solvent and a co-solvent mixture, predict whether solubility increases or decreases | SMILES + mixture composition | A / B | — | ~16.7% |
| `task4` | Pairwise compound comparison | Given a solvent and two compounds (SMILES), select the one with higher solubility | Solvent + 2 SMILES | A / B | 3000 | 50.0% |

---

## Usage

```python
from datasets import load_dataset

# Load a specific task
ds = load_dataset("levakrasnovs/SoluBench", "task1")
print(ds["test"][0])

# Load all tasks
for task in ["task1", "task2", "task3", "task4"]:
    ds = load_dataset("levakrasnovs/SoluBench", task)
    print(f"{task}: {len(ds['test'])} rows")
```

---

## Data fields

### Task 1 — Pairwise solvent comparison
| Field | Description |
|-------|-------------|
| `id` | Unique row identifier |
| `DOI` | Source publication |
| `SMILES` | Compound SMILES |
| `Temperature_K` | Temperature in Kelvin |
| `Solvent_A` | First solvent |
| `Solvent_B` | Second solvent |
| `LogS_A` | Experimental log solubility in solvent A (mol/L) |
| `LogS_B` | Experimental log solubility in solvent B (mol/L) |
| `Delta_LogS` | \|LogS_A − LogS_B\| |
| `Answer` | Correct answer (A or B) |

### Task 2 — Best solvent selection
| Field | Description |
|-------|-------------|
| `id` | Unique row identifier |
| `DOI` | Source publication |
| `SMILES` | Compound SMILES |
| `Temperature_K` | Temperature in Kelvin |
| `Best_solvent` | Solvent with highest solubility |
| `Best_LogS` | LogS in best solvent |
| `Second_best_solvent` | Solvent with second highest solubility |
| `Second_best_LogS` | LogS in second best solvent |
| `Delta_LogS_best_minus_second` | LogS difference between best and second best |
| `All_solvents` | Semicolon-separated list of all solvents |
| `All_LogS` | Semicolon-separated list of all LogS values |
| `Answer` | Correct answer (letter corresponding to best solvent) |

### Task 3 — Co-solvent effect prediction
| Field | Description |
|-------|-------------|
| `id` | Unique row identifier |
| `DOI` | Source publication |
| `SMILES` | Compound SMILES |
| `Temperature_K` | Temperature in Kelvin |
| `Base_solvent` | Pure base solvent |
| `Added_solvent` | Co-solvent added to the mixture |
| `Fraction_base` | Fraction of base solvent |
| `Fraction_added` | Fraction of added co-solvent |
| `New_composition` | Human-readable mixture description |
| `Fraction_type` | Type of fraction (mole/volume/mass) |
| `Delta_LogS_new_minus_base` | LogS(mixture) − LogS(pure base) |
| `Answer` | A = solubility enhanced, B = solubility reduced |

### Task 4 — Pairwise compound comparison
| Field | Description |
|-------|-------------|
| `id` | Unique row identifier |
| `Solvent` | Solvent name |
| `Temperature_K` | Temperature in Kelvin |
| `SMILES_A` | SMILES of compound A |
| `SMILES_B` | SMILES of compound B |
| `Compound_name_A` | Name of compound A |
| `Compound_name_B` | Name of compound B |
| `LogS_A` | Experimental log solubility of compound A (mol/L) |
| `LogS_B` | Experimental log solubility of compound B (mol/L) |
| `DOI_A` | Source publication for compound A |
| `DOI_B` | Source publication for compound B |
| `delta_logS` | LogS_A − LogS_B |
| `Answer` | Correct answer (A or B) |

---

## Source datasets

| Dataset | Paper | Download |
|---------|-------|----------|
| BigSolDB v2.0 | [Nature Scientific Data](https://www.nature.com/articles/s41597-025-05559-8) | [Zenodo](https://doi.org/10.5281/zenodo.15094979) |
| MixtureSolDB | [ChemRxiv](https://doi.org/10.26434/chemrxiv-2025-m51v8) | [Zenodo](https://doi.org/10.5281/zenodo.18660057) |

---

## Model results

Full model evaluation results (20+ LLMs including GPT, Claude, Gemini, Grok, Qwen, DeepSeek, GLM) are available in the [GitHub repository](https://github.com/levakrasnovs/SoluBench/tree/main/results).

---

## Citation

> Citation will be added upon preprint publication.

---

## License

[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)