Today, we are incredibly excited to present LiteMol-1, our very first foundation model from LiteFold. Structure-based models like BoltzGen, O-Design, and RFdiffusion have become the de facto standard for designing biomolecules. However, they are expensive to run at scale. In many campaigns, we have to generate tens of thousands of designs and rigorously filter them down to a handful of top candidates. More importantly, most molecule design systems are primarily optimized around binding. But what about everything else that makes a molecule a useful therapeutic: ADME, toxicity, drug-likeness, membrane permeability, synthesizability, selectivity, and more? That’s where we introduce LiteMol-1, our first Multi-Molecule Foundation Diffusion Language Model, pre-trained from scratch. With a single set of weights, LiteMol-1 can conditionally generate small molecules, peptides, cyclic peptides, depsipeptides, peptides with ncAAs, macrocycles, and PROTACs. You can generate molecules unconditionally or condition generation on a protein target sequence. You can in-paint molecules, preserve parts of an existing scaffold, generate non-canonical peptides, cyclize designs, and continuously edit and regenerate them. We also introduce a Monte Carlo Tree Search framework for multi-objective molecular design. Instead of trying to make one molecule perfect at everything, the search keeps a set of promising molecules, each with different trade-offs across properties. This matters because molecular design is fundamentally not a single-objective optimization problem. Finally, and probably the part we are most excited about: this is a model for Agents. LLMs are not particularly efficient interfaces for repeatedly reasoning over thousands of large PDB/CIF files inside an AutoResearch loop. Sequence space is different. SMILES and molecular sequences are compact, editable, and much easier for an agent to inspect, compare, modify, and reason over. So we treat LiteMol-1 as an infinite molecular canvas. The model generates possibilities. The agent takes inspiration from them, evaluates them, edits them, optimizes them, and generates again. The AutoResearch loop continues until it finds candidates that satisfy the given design objectives; while bringing in expensive structure prediction, docking, or simulation only when they are actually needed. Our evaluations across peptide and small-molecule generation show that LiteMol-1 is competitive with, and in several settings on-par with or better than, frontier structure-based and sequence-based models, while operating at a fraction of the generation cost and time. And this is only the first step. Check out our technical research blog post in the comments to learn more about LiteMol-1.

Less Time Connecting the Dots Research generates data. But how much time does your team spend connecting it? Structure predictions, screening results, molecular analysis and experimental findings often sit across different workflows. LiteFold aims to bring scientific intelligence and computational workflows closer together, helping researchers focus more on insights and less on fragmented analysis.

Unified Scientific Context Layer

These are all xls / csv files, which renders in real time

And modality agnostic bio-molecule design

If you wanna try such a similar system, an autonomous system of agents working together to understand the disease biology, finding targets, designing binders, optimizing them, ensemble models and a lot more! Try at studio.litefold.ai

Don't give your AI hard time to interpret your lab results by imploding their context.

Another heavily detailed report dropping soon

We are a year old 🐥

Its a good day, releasing everything on proteinbase soon

Cheers to one year at LF!

Oct 25/2025 vs Oct 28/2026 From no knowledge of what a pocket is to some knowledge now! From nothing to our own platform, to making our own models, working with companies, and keeping in track of our own experiments. Own the intelligence! Cheers to a year at LF!

Thanks! To clarify, ipSAE scores residues with PAE less than a generous cutoff (choose 10 or 15 Å), and then scales the PAEs by the TM formula (proportional to cube root) on the number of residues that pass the PAE cutoff in the scored chain (not just those in the interface).

I think a huge part at LiteFold internally went in finding the right order of evaluations and choosing the right metrics and their cutoff. The good part is, all those research are baked inside the LiteFold's BioDesign Platform. So if you want to engineer molecules with all the best practices, you know where to find us!

High ipTM. Low ipSAE. Should you order the peptide you designed? ipTM reflects global confidence in chain placement, but it's normalized over total chain length, so a high score doesn't guarantee a well-resolved binding interface. ipSAE focuses on inter-chain residue pairs with low predicted alignment error and scales confidence based on the interface, rather than the full chain length and is computed per direction (A→B, B→A) to expose asymmetric confidence. Score ipSAE for your peptides using the (github.com/DunbrackLab/IPSAE ) or on LiteFold.

My eyes are better now

As said before, if we are going to do agentic science, lets do it in a right, controlled way, just like how controlled reproducible experiments looks like. All u gotta do is just connect ur lab.