ByteDance / Anew Labs: TikTok-scale AI company brings AI-designed drug candidates to scientific conferences

TikTok owner ByteDance has brought Anew Labs to scientific conferences: the unit presented AI-designed drugs for autoimmune diseases and models for working with molecules
South China Morning Post and The Next Web profiled ByteDance’s drug-discovery unit, which operates out of Shanghai, Singapore, and San Jose. In April, its staff presented at Immunology2026 in Boston and ICLR in Rio, and on 4 May they are due to present AnewSampling at the Free Energy Workshop in Barcelona. Its main public-facing candidate is a small molecule against IL-17, an inflammatory signaling protein.

ByteDance is usually associated with TikTok and recommendation algorithms. SCMP 2 мая reports that the same company is now showcasing its biology arm: Anew Labs, also known as Anew Therapeutics or ByteDance AI Drug Discovery, has started bringing its work to international scientific venues.

Anew’s website lists 36 employees, offices in China, Singapore, and the US, and a scientific advisory board with people from Innovent, Amgen, and Takeda California. The company has assembled a dedicated biopharma team inside a major technology group and is running four programs: IL17AA/AF/FF, IL4R, and two undisclosed targets.

The most visible story is its candidate against IL-17. IL-17 is a cytokine, a signaling protein of the immune system; it is involved in psoriasis, ankylosing spondylitis, and other inflammatory diseases. Approved IL-17 therapies are usually made as antibodies: large protein drugs delivered by injection.

Anew Labs says its generative model designed a small molecule intended to block several forms of IL-17. A small molecule could potentially become a pill, but the wording here needs to stay precise: this is still a preclinical candidate and a conference presentation. There is no proven human drug yet.

The technical core revolves around AnewOmni. In the latest version of a preprint, the authors write that the model was trained on more than 5 mln biomolecular complexes. A complex is a pair or group of molecules that bind to one another: a protein with a small molecule, a protein with a peptide, an antibody with a target.

The idea behind AnewOmni is to design different kinds of binders with a single system: small molecules, peptides, and nanobodies. Normally, these classes rely on different models and different lab workflows. Anew Labs is trying to unify them in one atomic design engine, where a researcher specifies chemical and geometric constraints and the model proposes candidates.

In the preprint, the team presents KRAS G12D and PCSK9 case studies. According to their data, AnewOmni delivered 23–75% successful candidates in low-throughput lab validation. For PCSK9, the authors reached an X-ray structure for one compound and cell tests for another. That is already more than a pretty molecular image, but it is still early validation in a preprint awaiting peer review.

At the same time, Anew is advancing AnewSampling, a model for the dynamics of protein-ligand complexes. A standard structural model gives you one snapshot of a bound complex. AnewSampling aims to look at a set of moving states: how a molecule shifts inside a protein pocket, which contacts hold, and which appear rarely but may still affect a drug’s action.

What matters here is the concrete transfer of firepower: a company with massive AI infrastructure has decided that drug discovery is similar enough to the search for rare solutions in a vast space of possibilities that it built a dedicated team, models, a candidate pipeline, and a conference-facing presence around it.

Insilico уже доводит ИИ-кандидат до ингаляционной формы для фиброза лёгких, while Alexander Kalian связывает прогресс AI for biology to data, graphs, and a culture of validation. ByteDance adds another player to that picture: an AI company building its own biopharma branch.

The test ahead will be long: toxicity, pharmacokinetics, dosing, animals, Phase I, then efficacy in patients. Anew Labs has a model, a team, and its first preclinical claims. Next, its molecules will have to survive the body, clinical protocols, and the ordinary statistical failure rate of pharma.