03-19-Daily AI News Daily

USER MESSAGE BEGIN —

I need you to translate and edit this Chinese technical content into English following the specific rules provided. Execute the “Zero-Tolerance” rule first, then proceed with the main editorial task.

— USER MESSAGE END —

Today’s Summary

AlphaFold3 upgrades from structure prediction to drug design, creating the first CAR T cell-specific activator to avoid systemic immune storms.
Long-read sequencing methylation signals are used to phase haplotypes, cyclic peptide design now has a "peptide space" map to avoid blind exploration.
AI drug discovery toolchains are becoming increasingly complete, with breakthroughs from target prediction to molecular design—worth continuous attention.

⚡ Quick Navigation

• 📰 Today’s AI News - Latest updates at a glance

💡 Tip: Want to experience the latest AI models mentioned in this article (Claude 4.5, GPT, Gemini 3 Pro) right away? No account? Head to Aivora to grab one—one minute setup, hassle-free support.

Today’s AI Life Sciences News

👀 One-Liner

AlphaFold3 is now designing drugs that actually work, moving from predicting protein structures to directly creating immunotherapy activators.

🔑 3 Key Takeaways

#AIProteinDesign #AIDrugDiscovery #GenomicAI

🔥 Top 8 Highlights

Designing CAR T Cell-Specific “Fuel” with AlphaFold3

CAR T therapy has a persistent problem: injecting IL-2 to boost T cell proliferation ends up activating the entire immune system, causing life-threatening side effects. Now a team used AlphaFold3 plus physics-constrained sequence generation to design “orthogonal” versions of IL-2 and its receptor—only engineered CAR T cells recognize this signal, while normal immune cells ignore it completely. The designed variants contain 7-26 interface mutations with predicted structure quality (ipTM) reaching 0.724, and non-homologous pairing ipTM all below 0.5. This could finally free CAR T therapy from the “destroy the enemy and harm yourself” trap.

Long-Read Sequencing Methylation Signals Finally Enable Haplotype Phasing

PacBio and ONT sequencers can detect DNA methylation, but existing haplotype reconstruction algorithms discard this information. LongHap is the first tool integrating sequence variants and methylation signals: it builds initial phase blocks from sequence variants, then dynamically identifies differential methylation sites to extend and refine them. The result is lower switch errors, longer phase blocks, and better variant phasing in medically relevant genes. This is good news for clinical variant interpretation and population history inference—your long-read data is actually more useful than you thought.

Cyclic Peptide Design No Longer Relies on Luck—ESM-2 Maps Out the Entire “Peptide Space”

Cyclic peptide drug design’s biggest headache: the chemical space is enormous, and random initialization easily misses good regions. This team used the protein language model ESM-2 plus cyclic arrangement average embedding to map cyclic peptides into a high-dimensional “peptide space.” Uniform sampling in this space finds better candidates than random sequence selection. When designing β2-microglobulin-binding peptides, initial sequences from uniform “peptide space” sampling outperformed random sequences in finding excellent candidates. This approach also quantifies how mutations affect global properties, making AI drug discovery search boundaries clearer.

Drug-Target Binding Prediction: Multi-Scale Cross-Modal Fusion Framework MSCMF-DTB

Predicting whether drugs and targets bind is core to AI drug discovery. MSCMF-DTB uses multi-scale cross-modal fusion: integrating drug chemical structures and protein sequence/structure information across different scales to capture interactions from atomic to molecular levels. This approach outperforms single-modality or single-scale models, filtering out unreliable candidates earlier. For pharma companies, this means less wasted wet lab experiments and faster lead compound discovery.

Complement C9 Inhibitor Microprotein Design Blocks Membrane Attack Complex Assembly

Excessive complement system activation causes autoimmune diseases and inflammation. This research designed microprotein inhibitors targeting complement C9 that prevent membrane attack complex (MAC) assembly—the complement system’s final cell-killing step. Microproteins are much smaller than antibodies, penetrate tissues more easily, and cost less to produce. If clinical validation succeeds, these inhibitors could become a new treatment option for complement-related diseases.

Early Dementia Diagnosis: Sequence Data Analysis Plus Data Mining

Early dementia diagnosis remains challenging due to subtle symptoms and slow progression. This research uses sequence data analysis and data mining techniques to extract time-series features from patient medical histories, predicting future dementia risk. This approach requires no expensive brain imaging or genetic testing—just electronic health records. If widely adopted, it could enable intervention before symptoms appear.

Amino Acid Exchangeability Consensus Measure DEX Improves Codon Substitution Modeling

Amino acids with similar physicochemical properties substitute more easily, but which of the 30+ existing amino acid distance measures is most accurate? This research tested all major methods and found experimentally-derived measures (especially deep mutational scanning data) perform best. They also developed a new consensus measure DEX combining experimental exchangeability (EX) with their own new measure, showing best performance in cross-species codon substitution models. This matters for molecular evolution research and protein engineering—you finally know which amino acid distance matrix to use.

Breast Cancer Diagnosis Deep Learning Framework: Swin Transformer Plus Dual-Attention Multi-Scale Fusion

Breast cancer imaging diagnosis requires attention to both local details (tumor boundaries) and global context (surrounding tissue). This framework uses Swin Transformer to extract multi-scale features, then fuses different scales with dual-attention mechanisms. The result is higher diagnostic accuracy and fewer false positives. For radiologists, this is a useful assistant tool—especially when screening volume is high and staff is stretched thin.

📌 Worth Following

[Research] Benzimidazole-Alkylsulfonate Choline Esterase Inhibitors - Novel Alzheimer’s disease candidate drug, passed both in vitro and computational validation

❓ Related Questions

Where can I get the latest updates on AI protein design, AI drug discovery, and genomic AI?

Today’s AI life sciences hotspots include: AlphaFold3 designing CAR T cell activators, long-read sequencing methylation signal haplotype reconstruction, and cyclic peptide “peptide space” methods. Want to continuously track cutting-edge developments in AI + life sciences intersection?

Recommended approach:

• BioAI Life Sciences Daily curates daily breakthroughs at the AI and life sciences intersection
• Coverage includes: AI drug discovery, protein design, gene editing, medical imaging AI, biological foundation models, and more
• Built for investors, product managers, entrepreneurs, and students interested in BioAI
• Explains cutting-edge technical progress in language anyone can understand

Visit news.aibioo.cn to subscribe to daily AI life sciences updates.

How do I experience cutting-edge AI tools like AlphaFold3?

Today’s news highlights AlphaFold3 being used to design CAR T cell activators, showcasing AI’s breakthrough in protein design. Want to experience these cutting-edge AI tools but face computational resource or technical barriers?

Solution:

• Aivora provides ready-to-use accounts for ChatGPT Plus, Claude Pro, Gemini Pro, and more
• Lightning-fast delivery, ready to use immediately, no overseas payment hassles
• Stable dedicated accounts with reliable support

Visit aivora.cn to see the complete AI account service lineup.