冷暖人生

China’s top pork-producing province rolls out homegrown 'gene chip' for breeding quality pigs_我的网站

A |     　　当地时间8月20日，谷歌宣布，其开源端侧模型家族Gemma的总下载量已突破10亿次。这款由谷歌DeepMind推出的开源大模型，自发布以来，全球开发者基于它衍生出超10万个定制变体，形成名为“Gemmaverse”的开源生态。
　　谷歌列举了多个Gemma家族应用场景：从太空本地图像分析、医疗报告标准数字化、癌症创新治疗方案再到动物语言识别，Gemma家族模型正实实在在地应用到各个垂直行业。

B | 　　其中，以Gemma为基座的模型在生物医药科研领域产出了多项标志性成果。

C |     

    Yuzhu-1 Henan indigenous pig 15k breeding array Photo: Courtesy of Institute of Animal Husbandry and Veterinary Science of Henan Academy of Agricultural Sciences
China's largest pork-producing province has introduced a genomic breeding tool designed for indigenous pig breeding, a sign that the province is shifting from expanding volume toward improving hog quality.
The Henan Academy of Agricultural Sciences announced the launch of Yuzhu-1 Henan indigenous pig 15k breeding array - commonly known as a gene chip - developed by the pig breeding team at its Institute of Animal Husbandry and Veterinary Science. It is the first such tool customized for the province's native pig stock, the institute said on Monday.
Henan is home to multiple native pig breeds prized by consumers for their good flavor. However, they carry a commercial burden compared with imported lean-meat breeds, as they grow more slowly, consume more feed per kilogram of weight gain and deliver thinner margins, noted Xing Baosong, head of the pig research office at the institute.
Lifting those hurdles depends on sharper selection, which in local herds has long relied on visual assessment of physical traits, which is slow and imprecise. Genomic tools can speed things up, but mainstream commercial chips were built on imported breeds and lose accuracy on native Chinese stock. Yuzhu No 1 instead draws on whole-genome resequencing data from five of Henan's native pig breeds and two locally bred lines.
"In breeding we rely on estimated breeding values to judge an animal's genetic potential," Wang Ligang, a professor at the Institute of Animal Sciences under the Chinese Academy of Agricultural Sciences, told the Global Times on Monday. "An array matched to the breed makes that estimate more accurate. Previously there was a real chance we were selecting the wrong animals. Now we are far more likely to be selecting the right ones - and to select accurately for the traits a breed is known for, whether that is tender meat or high fertility."
The efficiency gap between native and imported pig breeds remains wide, said Wang, who is also with the Pig Branch of the Chinese Association of Animal Science and Veterinary Medicine. Feed conversion ratios for local Chinese pigs run around 3.0, against below 2.5 for imported breeds, while reaching 100 kilograms takes native pigs 180 to 190 days compared with 130 to 140 days for exotic lines, he said. 
Despite this, China's native pig breeds have their advantages.
"Our pigs taste better, partly because they are raised longer," Wang said. "And they are far more robust. Exotic breeds can suffer heat stress and die in summer; when disease hits, they can go down in large numbers. Local pigs rarely do. They tend to recover. Some breeds in Northeast China can even tolerate severe cold."
He added that imported breeds drive efficiency, while native breeds ensure quality and offer a reservoir of genetic diversity for disease and stress resistance - traits that can be bred back into commercial lines.
The province slaughtered 64.74 million pigs in 2025, up 7.4 percent year on year, with pork output at 5.22 million tons, according to the provincial statistics bureau, making it the top pork and pig producer nationwide, according to industry statistics. 
In the first half of 2026, pig slaughter in the province rose a further 6.4 percent, accelerating 5.3 percentage points from the first quarter, against national growth of 1.7 percent. Xing said that more high-quality local black pigs will reach the market as Yuzhu No 1 is applied more broadly. 
Other major pig production provinces such as Southwest China's Sichuan, Northwest China's Shaanxi and South China's Yunnan have all developed their own arrays.
"Provinces with the most developed pig production are the ones pushing upmarket," Wang said. "Scale is no longer the only measure. If every province reaches this level, a pattern of one dedicated array per province could gradually take shape for China's native breeds."
。谷歌一方面推出面向医疗场景的垂直模型MedGemma，支撑医疗报告结构化、医学影像辅助分析等应用；另一方面联合科研机构，依托Gemma底座攻坚癌症机制研究。

D | 　　其中最受行业关注的是谷歌DeepMind、谷歌研究院与耶鲁大学合作开发的Cell2Sentence-Scale 27B（C2S-Scale）模型。　　传统的癌症疫苗治疗思路，是让免疫系统通过识别身份标签来识别敌人。但这面临一个“瞄准困境”：一般来说，肿瘤的“身份标签”取自肿瘤本身,但很多肿瘤细胞表面缺少能被免疫T细胞识别的、独特的“身份标签”（即肿瘤特异性抗原），同时，它们还主动穿上“防护服”（如高表达PD-L1蛋白），成为所谓的“冷肿瘤”。结果就是，对于“冷肿瘤”，免疫系统即使大军压境，也像得了“脸盲症”，看不见敌人，或者看见了也被强行抑制，导致治疗无效。　　针对免疫治疗难以起效的“冷肿瘤”难题，C2S-Scale按活跃程度排序，将单细胞基因表达数据转化为类文本的“基因句子”，让大语言模型能直接“读”懂细胞。随后，该模型完成4000余种药物的虚拟筛选，预测出silmitasertib联合低剂量干扰素的用药组合。　　C2S-Scale锁定的候选药物中，仅10%-30%在过往文献中被提及，其余与筛选目标无已知关联。最终被选中的是silmitasertib（CX-4945），一种CK2激酶抑制剂。　　这一AI提出的全新治疗假说，后续在人源活细胞实验中得到验证，研究团队用模型训练时从未见过的人类神经内分泌细胞进行实验，结果是单独用药无效果，单用低剂量干扰素只有轻微提升，两者联用后抗原呈递提升约50%，证实联用方案可显著提升抗原呈递水平，属于少数AI产出全新癌症治疗机制并获得实验室验证的案例，有望提升肿瘤细胞抗原呈递能力，把冷肿瘤转化为对免疫治疗敏感的热肿瘤。　　该模型相关成果已对外开源，可供全球科研团队继续拓展研究。需要注意的是，该发现目前仅停留在体外细胞实验阶段，尚未进入临床。　　随着下载量冲高，Gemma证明开源大模型不再局限对话生成，能够直接介入原始生物数据挖掘，深度探索癌症治疗新机制，为AI医药科研提供低成本的技术底座。（文章来源：科创板日报）。

Current article:http://xxxusio.ruwentaqintiaozhupapaigai.shop/qg7h6/20260826/240862.html

Published on:08:25:39