Multi-Modal Foundation · mist-models
MIST 1.8B · HIV
A 1.8-billion-parameter molecular foundation model predicting HIV-1 replication inhibition from SMILES strings.
What MIST 1.8B · HIV is
MIST 1.8B · HIV is a molecular property prediction model fine-tuned to classify whether small molecules inhibit human immunodeficiency virus type 1 (HIV-1) replication. Built by the Electrochemical Energy Group at the University of Michigan, Ann Arbor, the model is part of the Molecular Insight SMILES Transformers (MIST) family of foundation models for chemical space exploration.
The model operates on chemical structures represented as Simplified Molecular Input Line Entry System (SMILES) strings. It was initialised from a 1.8-billion-parameter encoder pre-trained on the Enamine REAL Space database and subsequently fine-tuned on the MoleculeNet HIV benchmark dataset, which compiles experimental results for compounds tested in human T-cell lines.
By predicting antiviral activity directly from molecular sequences, the model aids medicinal chemists and computational pharmacologists in evaluating and prioritising prospective drug candidates without requiring immediate wet-lab assays.
How it works
The underlying backbone is a RoBERTa-PreLayerNorm encoder featuring 24 layers, 18 attention heads, a hidden size of 2,304, an intermediate size of 9,216, and a context window of 2,048 tokens. The base model was pre-trained using masked language modelling with cross-entropy loss on Kekulized SMILES strings from Enamine REAL Space, an extensive collection of synthetically accessible molecules.
For HIV activity prediction, the encoder is coupled with a task-specific classification head comprising a two-layer multi-layer perceptron (MLP) with Gaussian Error Linear Unit (GELU) activations and dropout. Hidden states are pooled using the representation of the first token in the sequence. Fine-tuning was carried out using binary cross-entropy loss and the AdamW optimiser on 41,127 compounds split 80/10/10 by molecular scaffold.
Key facts
- Architecture
- RoBERTa-PreLayerNorm encoder (24 layers, 18 heads) with two-layer MLP task head
- Parameter count
- 1.8 billion
- Pre-training data
- Enamine REAL Space
- Fine-tuning data
- MoleculeNet HIV dataset (41,127 compounds)
- Context length
- 2,048 tokens
- Input format
- Kekulized SMILES strings
- Output format
- Binary classification logits or pooled feature vectors
What you can use it for
- HIV-1 Inhibitor Screening
- Evaluate small molecules to predict whether they inhibit HIV-1 replication in human T-cell lines.
- Chemical Library Prioritisation
- Score and rank large collections of candidate SMILES strings before committing to in vitro validation.
- Molecular Feature Extraction
- Derive dense vector embeddings from SMILES strings using the pre-trained and fine-tuned encoder for downstream computational chemistry tasks.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
MIST 1.8B · HIV
Strengths and limits
Good at
- Pre-trained on Enamine REAL Space, grounding representation learning in synthetically accessible chemical space.
- Evaluated using a scaffold split rather than random partitioning, providing a realistic test of generalisation to unseen core structures.
- Supports long input lengths of up to 2,048 tokens, enabling the representation of larger or structurally intricate molecules.
Watch out for
- Requires a Rust installation to compile and run the prerequisite SMIRK tokenizer.
- Outputs only binary classification for anti-HIV activity rather than detailed quantitative affinity or potency values.
- Requires Kekulized SMILES strings as inputs to match the training data formatting.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use MIST 1.8B · HIV commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run MIST 1.8B · HIV without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on 1× A100 16GB, typically in 12s, and a run costs about 70 credits. Visitors without an account get 1,000 credits a month.
What input does MIST 1.8B · HIV accept?
The model takes molecular structures formatted as SMILES strings, specifically Kekulized SMILES strings, tokenised with the SMIRK tokenizer.
What dataset was used to fine-tune the model?
It was fine-tuned on the HIV dataset from the MoleculeNet benchmark, which contains 41,127 compounds tested for inhibition of HIV-1 replication in T-cell lines.
What metrics and split strategies were used during testing?
Performance was evaluated using Area Under the Receiver Operating Characteristic (AUROC) on an 80/10/10 scaffold split of the MoleculeNet HIV dataset.
What hardware was used to train the model?
Pre-training was conducted on 32 NVIDIA A100-SXM4-80GB GPUs over approximately 7 days, whereas fine-tuning was completed on a single NVIDIA A100 GPU.
How to cite it
@online{MIST, title = {Foundation Models for Discovery and Exploration in Chemical Space}, author = {Wadell, Alexius and Bhutani, Anoushka and Azumah, Victor and Ellis-Mohr, Austin R. and Kelly, Celia and Zhao, Hancheng and Nayak, Anuj K. and Hegazy, Kareem and Brace, Alexander and Lin, Hongyi and Emani, Murali and Vishwanath, Venkatram and Gering, Kevin and Alkan, Melisa and Gibbs, Tom and Wells, Jack and Varshney, Lav R. and Ramsundar, Bharath and Duraisamy, Karthik and Mahoney, Michael W. and Ramanathan, Arvind and Viswanathan, Venkatasubramanian}, date = {2025-10-20}, eprint = {2510.18900}, eprinttype = {arXiv}, eprintclass = {physics}, doi = {10.48550/arXiv.2510.18900}, url = {http://arxiv.org/abs/2510.18900}}
Written from MIST 1.8B · HIV's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.