Anew Labs

Anew Labs (Chinese: 新生实验室) is a Shanghai-headquartered AI drug discovery company that began as the internal drug discovery team of ByteDance and was spun out as a separate company in 2026.^[2][10] ByteDance's AI drug discovery team was formed in 2021 under Liu Kai (Kai Liu), who is now the company's chief executive.^[7][15] On 16 September 2026 Reuters reported, citing two people with knowledge of the matter, that Anew Labs had completed a $290 million first external fundraising at a valuation of $1.5 billion, with ByteDance keeping a 56% stake. The round was led by HSG (formerly Sequoia China), IDG Capital and Hillhouse Investment, with 5Y Capital as co-lead.^[2] Anew Labs develops its own AI models: AnewFold for structure prediction, AnewSampling for conformational dynamics, AnewOmni for generative binder design, AnewDesign for antibody design and AnewMind, a reasoning large language model for drug discovery decisions. It also runs a pipeline of four disclosed preclinical programs. The most advanced is an oral small-molecule inhibitor of the IL-17A and IL-17F cytokine dimers.^[1][18] On 25 September 2026 Endpoints News named the company to its annual Endpoints 11 list of biotech startups.^[15][16]

Overview

History

Origins inside ByteDance

Chinese reports say ByteDance's AI Lab began hiring for AI drug discovery at the end of 2020. A dedicated AI drug discovery team was then formed in 2021, led by Liu Kai and made up of AI-for-science algorithm researchers and experienced pharmaceutical scientists.^[7][9] Before joining ByteDance, Liu spent seven years as a healthcare investor. A June 2026 profile reprinted by Sina Finance says he began at IDG Capital in 2014, joined Fire Stone Investment (火山石投资) as a managing director in 2016 and joined ByteDance in 2021 as head of AI drug discovery.^[3][11] IDG Capital, his former firm, is one of the lead investors in Anew Labs' 2026 round.^[3]

The team released a stream of research under ByteDance and Anew names before the spin-off. Work the company now lists as its own includes scPertBench, a benchmark of AI models for in silico gene perturbation (preprint posted in December 2024), and ImmunoSeq, an antibody immunogenicity predictor whose August 2025 preprint already carries both a ByteDance affiliation and Anew Therapeutics Pte. Ltd. of Singapore, with correspondence at ByteDance addresses.^[1][34][37] In March 2026 the team published AnewSampling and AnewOmni. The AnewOmni paper was a collaboration with Tsinghua University.^[6][25][27]

The company also counts two ByteDance structure prediction projects in its model suite. One is Protenix, an open-source reproduction of AlphaFold 3. The other is SeedFold, a model for studying how folding models scale.^[23] Both began as ByteDance projects. The Protenix code is hosted under ByteDance's GitHub organization, and its README credits the "ByteDance AML AI4Science Team" in its citation.^[31] Its README also directs collaboration inquiries to an anewbt_mind@bytedance.com address.^[31] Chinese media have described the organization in different ways. A June 2026 analysis by 氨基观察 (Amino Observation), carried by 21st Century Business Herald, said that besides Anew Labs ByteDance had two other internal AI drug lines, SeedFold and Protenix, each training its own models.^[39] Also in June, 36Kr's Intelligent Emergence (智能涌现) reported that the internal team responsible for protein structure prediction models had already been merged into Liu Kai's group.^[10] A 36Kr column published later that month said a separate ByteDance AI4S team, credited with Protenix and PXDesign, had been reorganized in May under technology vice president Yang Zhenyuan. The column concluded that the AI4S team and Anew Labs sat on different management lines but shared people and technology.^[12]

In 2026 the unit began presenting its work in public under the names Anew Labs and Anew Therapeutics. In mid-April Chris Li, head of biology, gave an oral presentation on the IL-17 program at the American Association of Immunologists' IMMUNOLOGY2026 meeting in Boston. The abstract lists the presenter's affiliation as "ByteDance AI Drug Discovery / Anew Therapeutics".^[18] In May, team members presented AnewSampling and the company's free-energy work at the Alchemistry free-energy workshop in Barcelona.^[6][30] The South China Morning Post reported at the time that the website listed 36 core team members. The same report named a scientific advisory board including Liu Yongjun, former president of Innovent Biologics; Ji Ma, a former principal scientist at Amgen; and Hua Zou, scientific director of protein chemistry at Takeda California.^[6]

Spin-off

On 10 June 2026 Intelligent Emergence reported that ByteDance's AI drug business had formally started a spin-off and independent fundraising process. The core team, algorithms, technology platform and existing pipeline assets would move into a new entity that ByteDance would still control. The report said the new company would keep drawing computing power from Volcano Engine, and that the core team numbered about 50.^[10] A person close to the spin-off told the outlet it was ByteDance's first attempt to commercialize AI for science, and that because biotech has its own industrial logic, independence would give the business more flexibility in decision-making.^[10] On 16 June a profile first carried by Z Finance and reprinted by Sina Finance reported that the new company was valued at about $1 billion, with a first round of close to $200 million under discussion. It said investors had not yet been finalized.^[11] The round that closed in September was larger than those June figures.^[2]

First external round

Reuters reported on 16 September 2026 that ByteDance had completed a $290 million fundraising for Anew Labs after spinning it off. The report cited two people with knowledge of the matter, and neither ByteDance nor the investors immediately commented.^[2] According to Reuters, one source said the round valued the company at $1.5 billion, and the second said ByteDance would keep 56%. One source said the spin-off was meant to support the unit's long-term development because AI drug discovery "follows a different industry logic and management approach" than ByteDance's core operations.^[2] Chinese outlets described the $1.5 billion figure as a post-money valuation (投后估值).^[7][8][9] The news page of the 生物医药产业国际合作大会, a biopharma partnering conference held in Shanghai, called it the largest AI drug discovery financing of the year, and 猎云网 (Lieyun) called it the largest single round in China's AI drug discovery sector.^[13][14]

Reports differ on which Hillhouse entity invested. Reuters, IPO早知道 (IPO Zaozhidao) and Yicai name Hillhouse Investment (高瓴投资).^[2][7][8] TNGlobal (TechNode Global), 瑞财经 (Rui Caijing) and Lieyun name its venture arm, GL Ventures (高瓴创投).^[4][9][14] Yicai lists only the leads, the co-lead and the two strategic investors (Sino Biopharmaceutical and the Shanghai Future Industry Fund).^[7] Endpoints News gives the company's total funding as $290 million and names HSG, IDG Capital and Hillhouse Investment as key investors.^[15] The round details rest on press reports citing unnamed sources; as of 1 October 2026 the news section of the company's website lists no funding announcement.^[1][2]

Leadership and organization

Endpoints News lists Kai Liu as chief executive.^[15] He is the corresponding author of the company's AnewDDE technical report.^[22] Named function heads in press coverage include Chris Li (head of biology), She Yuli (head of data) and Yu Haoyu (head of computational chemistry).^[6] Chinese reports put the core team at about 50 people. The SCMP said the company website listed 36 core members in May 2026.^[6][7][10]

The company's website says it has offices in Shanghai, San Francisco and Singapore. Its structured page metadata gives street addresses in Shanghai, Singapore and San Jose, California, and its job listings are for Shanghai, San Jose and Singapore.^[1][32] It describes itself as an organization where "drug discovery experts and AI researchers work side by side".^[1] Rui Caijing reported that the US office operates under the Anew Therapeutics name.^[9]

Research platform

On its homepage Anew Labs lists five platforms: AnewFold, AnewSampling, AnewOmni, AnewDesign and AnewMind.^[1] In September 2026 it combined several of them into AnewDDE, which it calls "an agentic Drug Discovery Engine".^[21][22]

AnewFold, Protenix and SeedFold

The company's structure prediction page groups three models: AnewFold, Protenix and SeedFold.^[23] It calls AnewFold its "in-house all-atom foundation model for biomolecular complexes". The company says AnewFold's clearest advantage is on novel targets that lack close structural analogs, and that it can locate allosteric and cryptic binding pockets.^[23] In the AnewDDE report the company says AnewFold reached top-1 antibody-antigen success rates of 62.3% on AF3-AB, 60.8% on PXMeter-AB and 76.2% on FoldBench-AB across five seeds. It reports 77.4% on a similarity-filtered FoldBench protein-ligand set and 71.8% on a post-cutoff molecular glue benchmark, against figures it gives as 45.0% for Protenix-v1 and 29.8% for Protenix-v2. It also reports an overall AUPRC of 0.751 for ligand-free pocket identification.^[21] All of these figures are self-reported.

Protenix is described on the same page as "the first model to comprehensively reproduce AlphaFold3". The Protenix README says the project's code and model parameters are released under the Apache 2.0 license.^[23][31] The README says Protenix-v1 was released on 5 February 2026 and Protenix-v2 on 8 April 2026. Protenix-v2 is a roughly 464-million-parameter model with gains on antibody-antigen prediction, and the README says its weights are proprietary and not released under any open-source license.^[31] SeedFold, posted to arXiv on 30 December 2025, studies how folding models should scale. Its authors report that widening the Pairformer works better than adding depth, introduce a linear triangular attention mechanism, and build a large distillation dataset. On that basis they report outperforming AlphaFold3 on most protein-related FoldBench tasks.^[33]

AnewSampling

AnewSampling is a generative model meant to produce the conformational ensembles of protein-ligand complexes that molecular dynamics simulations produce, at much lower cost. The company argues that a single static structure from co-folding models is insufficient for drug design.^[24] The preprint ("Learning the All-Atom Equilibrium Distribution of Biomolecular Interactions at Scale", first posted in March 2026) describes a quotient-space generative framework. It is trained on what the authors call the largest self-curated database of protein-ligand trajectories to date, with over 15 million conformations.^[25] The authors claim it is "the first model to faithfully reproduce MD at the all-atom level". They report that it outperforms prior generative methods on the ATLAS monomer benchmark and recovers coupled ligand and side-chain motions in CDK2 systems.^[25] The preprint lists its authors under two affiliations, ByteDance AI Drug Discovery and Anew Therapeutics, with Kai Liu and Haoyu Yu as corresponding authors.^[25]

AnewOmni

AnewOmni is an all-atom geometric latent diffusion framework. It represents amino acids and small-molecule fragments as "blocks", compresses them with a variational autoencoder, and runs an E(3)-equivariant diffusion model in the latent space.^[26] The preprint, whose corresponding authors are split between Tsinghua University and ByteDance, says the model was trained on more than 5 million biomolecular complexes.^[27] It reports designs of small molecules, peptides and nanobodies against the KRAS G12D switch II pocket, and of orthosteric peptides and allosteric small molecules against PCSK9. The preprint gives the experimental success rate across these campaigns as 23% to 75%. The company's page gives 29% to 75% for the KRAS case.^[26][27] For PCSK9, the company reports that a crystal structure of one designed compound matched the predicted pose to within 0.92 Å RMSD. It says another compound (Kd 2.98 µM) raised LDLR expression in cells at 100 µM to levels comparable with the clinical-stage reference compound AZD0780 at 300 µM.^[26]

AnewDesign

AnewDesign is an agent-assisted workflow for antibody and nanobody discovery, run through a natural-language interface the company calls AnewScience. In the workflow an AI agent gathers literature and structural context, proposes epitopes, launches diffusion-based design runs, filters candidates for sequence liabilities and diversity, and uses wet-lab measurements to guide the next round.^[28] The company says it builds on its team's earlier work on SeedProteo, PXDesign and Protenix-v2.^[28] In a proof-of-concept nanobody campaign against an internal target, the company reports these results:^[21][28]

The company notes that the measurements establish binding affinity only. It says developability and functional evaluation are later steps.^[28]

AnewMind

AnewMind is described as an internal LLM "at the hundred-billion-parameter scale", given full-parameter post-training for pharmaceutical knowledge, structure-property reasoning and multi-stage R&D decisions.^[29] In zero-shot ADMET and developability tests against frontier models, the company says AnewMind Preview ranked in the top five in all four aggregate settings and first on cyclic-peptide PAMPA permeability (Elo-AUROC 0.737).^[21] It also introduced PharmBench, an internal benchmark of 50 R&D scenarios and 200 questions written by pharmaceutical experts. On PharmBench the company describes AnewMind's performance as "competitive with leading frontier models worldwide"; the comparison covers 17 models.^[21][29] The company also reports AnewMind's scores on general benchmarks including MMLU-Pro and GPQA Diamond.^[29] As of October 2026 the website lists AnewMind as "Coming Soon".^[1]

AnewDDE and other tools

AnewDDE, the "agentic Drug Discovery Engine", is described in a technical report credited to the "Anew Labs Team" and dated 16 September 2026.^[22] It connects four components into a closed loop of hypothesis, ranking, candidate generation and experimental validation: AnewFold, AnewAffinity, AnewDesign and AnewMind.^[21] AnewAffinity is a binding-affinity ranking model with confidence estimates compatible with free-energy perturbation (FEP). On 199 ligands from eight retrospective JACS systems, the company reports a Pearson R² of 0.553 and a Spearman correlation of 0.724. It gives 0.486 and 0.620 for its own local evaluation of Boltz-2. Pairwise ΔΔG RMSE was 1.057 kcal/mol for AnewAffinity and 1.016 kcal/mol for Boltz-2. The company says each ligand-pair edge takes about 1.5 seconds on a single GPU.^[21] The SCMP reported the AnewDDE launch on 18 September and noted the company's description of it as a response to "fragmented" AI workflows.^[5]

Other published work listed on the company's research page includes:^[1]

Pipeline

The company's website shows four drug discovery programs, all before clinical development. The stages the website lists run from exploratory work through Hit ID, hit-to-lead and lead optimization to IND-enabling studies.^[1] As of 1 October 2026 the chart shows:

IL-17 program

IL-17A and IL-17F form three dimers (AA, AF and FF) that drive several autoimmune diseases. According to the IMMUNOLOGY2026 abstract, small molecules developed so far "have been restricted to IL-17A and lack activity on IL-17F".^[18] On 16 April 2026 in Boston, Chris Li presented orally bioavailable small-molecule inhibitors designed through "AI-enabled structure-based generative design and virtual screening".^[18] The abstract reports:^[18]

• Ki below 0.1 nM against IL-17A and below 10 nM against IL-17F in cellular assays.
• Activity on IL-17A more than 10-fold more potent than leading IL-17A-specific small molecules.
• Favorable oral bioavailability across multiple preclinical species.
• Dose-dependent suppression of IL-17F-driven cytokine production in vivo, with maximal inhibition comparable to the dual-specific antibody bimekizumab.

The authors called these "the first orally bioavailable small molecules with strong IL-17F engagement".^[18] More data followed at the FOCIS annual meeting in San Francisco in June 2026. The company says both talks were invited podium presentations.^[19] TONACEA, a Chinese pharmaceutical industry events company, wrote up the AAI talk and named the lead molecule AN-5162. It quoted a Ki of 6.2 nM for IL-17FF and below 0.1 nM for IL-17AA.^[20] All of these data are preclinical. The company's pipeline chart shows the program in lead optimization, with IND-enabling studies still planned.^[1][18]

Reception

Endpoints News named Anew Labs to its 2026 Endpoints 11, a list of 11 startups chosen from more than 150 nominations. The list was published on 25 September 2026.^[16] Endpoints titled its profile "A China-based rival to AI labs".^[15][16] The company's summary of the profile says Endpoints contrasted Anew's habit of publishing its research with Isomorphic Labs' more limited disclosure. According to that summary, Endpoints also said Anew had disclosed more pipeline progress than more heavily funded AI drug discovery companies such as Isomorphic and Xaira Therapeutics.^[17]

Commentary on the financing focused on the gap between the valuation and the early pipeline. In a 19 September post crediting TMTPost, the X account TechBuzzChina argued that a traditional pipeline valuation for four preclinical assets "might top out at $500 million". In its view the remaining roughly $1 billion of the valuation was a price on "the machine that produces pipelines". It described the 56% stake as a compromise between full ownership and an exit, and Sino Biopharmaceutical's investment as filling Anew's wet-lab, clinical and regulatory gaps.^[36] Some details in that post differ from other reporting. It calls Anew "roughly three years old", while Chinese outlets date the team to 2021.^[7][36] It also says the IL-4R program targets atopic dermatitis; the company's pipeline chart names only the target.^[1][36] Lieyun argued that the real value in AI drug discovery lies in pipelines and marketed drugs rather than in selling models and compute, and that the "technology parent plus independent entity" structure lets Chinese tech groups keep development flexibility while still drawing on the parent's technical base.^[14] The 36Kr column called both disclosed targets, IL-17 and IL-4R, iterative work on established targets rather than new-target discovery.^[12]

Context

Anew Labs is one of several AI drug discovery efforts tied to large technology companies. Isomorphic Labs, the Google DeepMind spin-off, announced a $2.1 billion Series B on 12 May 2026, led by Thrive Capital.^[5][35] Several outlets set the Anew round against Isomorphic's raise.^[5][7][36] Yicai framed big-tech approaches as two camps. In one, companies hold their own pipelines, as Isomorphic does and, in Yicai's account, as Anthropic began doing after acquiring Coefficient Bio in April 2026. In the other, companies provide platforms, such as NVIDIA's co-innovation lab with Eli Lilly and OpenAI's GPT-Rosalind.^[7] Among Chinese internet companies, Lieyun compared Anew with Baidu's incubation of BioMap (百图生科) and with Tencent's mainly investor and compute-provider role.^[14] The SCMP placed the round within a wider fundraising boom for Chinese AI drug developers in September 2026.^[5]

See also

• ByteDance
• ByteDance Seed
• AI drug discovery
• AI for Science
• Isomorphic Labs
• AlphaFold
• Protein folding
• Insilico Medicine

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21Anew Labs. "Anew Labs (新生实验室) | AI Drug Discovery & AI Pharmaceutical Company" (homepage: platforms, research, pipeline chart, teams, offices). Accessed 1 October 2026. anewbt.com
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Kane Wu (Reuters). "ByteDance completes $290 million fundraising for AI drug unit after its spin-off, sources say". 16 September 2026 (syndicated copy). regionalmedianews.com/...-its-spin-off-sources-say
3. ^1 ^2 ^3Ana-Maria Stanciuc. "ByteDance's AI drug unit Anew Labs raises $290M at $1.5B valuation, Reuters reports". The Next Web, 16 September 2026. thenextweb.com/...anew-labs-290m-ai-drug-discovery
4. ^TNGlobal. "ByteDance AI drug discovery spin-off Anew Labs raises $290M at $1.5B valuation". TechNode Global, 17 September 2026. technode.global/...-spin-off-anew-labs-raises-290m
5. ^1 ^2 ^3 ^4Xinmei Shen. "ByteDance spin-off Anew Labs raises US$290m as AI drug developers enjoy fundraising boom". South China Morning Post, 18 September 2026. scmp.com/...drug-developers-enjoy-fundraising-boom
6. ^1 ^2 ^3 ^4 ^5 ^6Coco Feng. "ByteDance's drug unit presents AI-designed therapies at global conferences". South China Morning Post, 2 May 2026. scmp.com/...-designed-therapies-global-conferences
7. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12吕倩 (Yicai). "字节拆出一家AI制药公司". 第一财经 via Sina, 16 September 2026. k.sina.cn/article_1702925432_6580947801901b53s
8. ^1 ^2IPO早知道. "字节跳动分拆的AI制药业务「新生实验室」完成2.9亿美元融资，估值达15亿美元". Via Tencent News, 16 September 2026. news.qq.com/...20260916A0BAOD00
9. ^1 ^2 ^3 ^4 ^5 ^6王敏 (瑞财经). "字节拆出一家AI制药公司Anew Labs，首轮融资2.9亿美元、投后估值15亿美元". 17 September 2026. m.rccaijing.com/news-7506189638047560780
10. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8三言科技 (citing 智能涌现). "字节跳动 AI 制药业务启动拆分独立融资 刘凯团队主导新公司". NetEase, 10 June 2026. 163.com/...KV2N5D8J0519U3I5
11. ^1 ^2鞭牛士 (reprinted from Z Finance). "字节拆分AI制药，10亿美元估值融资2亿，刘凯带队落地上海". Sina Finance, 16 June 2026. finance.sina.com.cn/...doc-inicqhix7765391.shtml
12. ^1 ^2"Talk AI：字节与Anthropic打响AI制药暗战". 36Kr, 26 June 2026. 36kr.com/...3869802108818695
13. ^生物医药产业国际合作大会 (CBPF). "字节AI制药'单飞'首秀：新生实验室融资2.9亿美元，字节仍控股56%". 24 September 2026. cbpfcn.com/...9305
14. ^1 ^2 ^3 ^4韩文静 (猎云网). "科技大厂，集体入局AI制药". Via Tencent News, 21 September 2026. news.qq.com/...20260921A0ADVY00
15. ^1 ^2 ^3 ^4 ^5 ^6Andrew Dunn. "Endpoints 11: ByteDance makes more than TikToks, with biotech spinout Anew Labs turning into a formidable China rival in AI bio". Endpoints News, 25 September 2026. endpoints.news/...s-a-china-based-rival-to-ai-labs
16. ^1 ^2 ^3Drew Armstrong. "Biotech's most exciting startups of 2026: Meet this year's Endpoints 11". Endpoints News, 25 September 2026. endpoints.news/...026-meet-this-years-endpoints-11
17. ^Anew Labs. "Anew Labs Named to the 2026 Endpoints 11". 25 September 2026. anewbt.com/...endpoints-11-2026
18. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Chris Li. "First Oral Small-molecule Inhibitors Designed by Generative AI Achieve Biologics-Like Pan-Inhibition of IL-17AA, AF, and FF Dimers". IMMUNOLOGY2026 (American Association of Immunologists), block symposium abstract, 16 April 2026. immunology2026.eventscribe.net/fsPopup.asp
19. ^Anew Labs. "2026 AAI Oral Presentation: First-in-Class Small-Molecule Pan-IL-17 Inhibitor". anewbt.com/...il17-aai-focis-2026
20. ^TONACEA. "ByteDance Unveils Its First AI-Pharma Pipeline: Tackling IL-17 with a 'Pan-Spectrum' Small Molecule". 28 April 2026. en.tonacea.com/...890
21. ^1 ^2 ^3 ^4 ^5 ^6 ^7Anew Labs. "AnewDDE: An Agentic Drug Discovery Engine". 19 September 2026. anewbt.com/...anewdde-agentic-drug-discovery-engine
22. ^1 ^2 ^3Anew Labs Team. "AnewDDE: An Agentic Drug Discovery Engine for Biomolecular Interaction Modelling and Closed-Loop Design" (technical report, PDF), dated 16 September 2026. anewlabs-research.tos-cn-beijing.volces.com/...port.pdf
23. ^1 ^2 ^3 ^4 ^5 ^6Anew Labs. "AnewFold" (research page covering AnewFold, Protenix and SeedFold). anewbt.com/...folding
24. ^1 ^2 ^3Anew Labs. "AnewSampling". anewbt.com/...anewsampling
25. ^1 ^2 ^3 ^4 ^5Wang, Y., et al. "Learning the All-Atom Equilibrium Distribution of Biomolecular Interactions at Scale". bioRxiv, 2026. biorxiv.org/...2026.03.10.710952v1
26. ^1 ^2 ^3 ^4 ^5Anew Labs. "AnewOmni". anewbt.com/...anewomni
27. ^1 ^2 ^3 ^4Kong, X., et al. "Programming Biomolecular Interactions with All-Atom Generative Model". bioRxiv, 2026. biorxiv.org/...2026.03.12.711044v1
28. ^1 ^2 ^3 ^4 ^5 ^6Anew Labs. "AnewDesign". anewbt.com/...design
29. ^1 ^2 ^3 ^4Anew Labs. "AnewMind". anewbt.com/...mind
30. ^1 ^2Anew Labs. "Anew Labs Showcases AI-Driven Free-Energy Research at the Alchemistry Workshop". 4 May 2026. anewbt.com/...fep-alchemistry-2026
31. ^1 ^2 ^3 ^4ByteDance. "Protenix: Protein + X" (GitHub README). Accessed 1 October 2026. github.com/...Protenix
32. ^Anew Labs. "Careers". anewbt.com/careers
33. ^Zhou, Y., et al. "SeedFold: Scaling Biomolecular Structure Prediction". arXiv:2512.24354, 30 December 2025. arxiv.org/...2512.24354
34. ^1 ^2Huang, Q., He, Y., Liu, K. "Antibody immunogenicity prediction and optimization with ImmunoSeq". bioRxiv, August 2025. biorxiv.org/...2025.08.14.670305v1
35. ^Isomorphic Labs. "Isomorphic Labs announces Series B investment round". 12 May 2026. isomorphiclabs.com/...es-series-b-investment-round
36. ^1 ^2 ^3 ^4TechBuzzChina (@TechBuzzChina). "ByteDance spins out AI drug unit at $1.5B valuation". X, 19 September 2026. x.com/...2101353138619879630
37. ^1 ^2Li, C., et al. "Benchmarking AI Models for In Silico Gene Perturbation of Cells". bioRxiv, 2024. biorxiv.org/...2024.12.20.629581v2
38. ^Anew Labs. "AnewSynth". anewbt.com/...anewsynth
39. ^张曦 (氨基观察). "当字节跳动开卷AI制药". 21st Century Business Herald (21财经), 23 June 2026. m.21jingji.com/...3a263c09b256b81a41bc5ec8665d570c
40. ^Anew Labs. "AnewDPSA". anewbt.com/...dpsa
41. ^Anew Labs. "scNext". anewbt.com/...scnext