2019年医学和生物学领域深度学习和神经网络十大基础研究进展

Nature—大脑思维解析和人工智能：科学家可依据人类皮层活动将大脑中的“语句”人工合成出来

英文摘要

Technologythat translates neural activity into speech would be transformative for peoplewho are unable to communicate as a result of neurological impairments. Decodingspeech from neural activity is challenging because speaking requires veryprecise and rapid multi-dimensional control of vocal tract articulators. Herewe designed a neural decoder that explicitly leverages kinematic and sound representationsencoded in human cortical activity to synthesize audible speech. Recurrentneural networks first decoded directly recorded cortical activity intorepresentations of articulatory movement, and then transformed theserepresentations into speech acoustics. In closed vocabulary tests, listenerscould readily identify and transcribe speech synthesized from corticalactivity. Intermediate articulatory dynamics enhanced performance even withlimited data. Decoded articulatory representations were highly conserved acrossspeakers, enabling a component of the decoder to be transferrable acrossparticipants. Furthermore, the decoder could synthesize speech when aparticipant silently mimed sentences. These findings advance the clinicalviability of using speech neuroprosthetic technology to restore spokencommunication.

参考文献：

Anumanchipalliet al (2019). Speech synthesis from neural decoding of spoken sentences. Nature.2019 Apr;568(7753):493-498.

NatBiotechnol—药物设计和人工智能：深度学习帮助科学家快速发现了DDR1激酶的抑制剂

英文摘要：

We havedeveloped a deep generative model, generative tensorial reinforcement learning(GENTRL), for de novo small-molecule design. GENTRL optimizes syntheticfeasibility, novelty, and biological activity. We used GENTRL to discoverpotent inhibitors of discoidin domain receptor 1 (DDR1), a kinase targetimplicated in fibrosis and other diseases, in 21 days. Four compounds wereactive in biochemical assays, and two were validated in cell-based assays. Onelead candidate was tested and demonstrated favorable pharmacokinetics in mice.

参考文献：

Zhavoronkovet al (2019). Deep learning enables rapid identification of potent DDR1 kinaseinhibitors. Nat Biotechnol. 2019 Sep;37(9):1038-1040.

Science—药物合成和人工智能：在人工智能计划的引领下，机器人可以自动完成部分有机化合物的合成过程

英文摘要：

The synthesis of complex organicmolecules requires several stages, from ideation to execution, that requiretime and effort investment from expert chemists. Here, we report a step towarda paradigm of chemical synthesis that relieves chemists from routine tasks,combining artificial intelligence-driven synthesisplanning and a robotically controlled experimental platform. Synthetic routesare proposed through generalization of millions of published chemical reactionsand validated in silico to maximize their likelihood of success. Additionalimplementation details are determined by expert chemists and recorded inreusable recipe files, which are executed by a modular continuous-flow platformthat is automatically reconfigured by a robotic arm to set up the required unitoperations and carry out the reaction. This strategy for computer-augmentedchemical synthesis is demonstrated for 15 drug or drug-like substances.

参考文献：

Coleyet all (2019). A robotic platform for flow synthesis of organic compoundsinformed by AI planning. Science. 2019 Aug 9;365(6453). pii: eaax1566.

NatMethods —小动物行为学研究和人工智能：深度神经网络能够快速评估和追踪小动物的姿势

英文摘要：

The need for automated andefficient systems for tracking full animal pose has increased with thecomplexity of behavioral data and analyses. Here we introduce LEAP (LEAPestimates animal pose), a deep-learning-based method for predicting thepositions of animal body parts. This framework consists of a graphicalinterface for labeling of body parts and training the network. LEAP offers fastprediction on new data, and training with as few as 100 frames results in 95%of peak performance. We validated LEAP using videos of freely behaving fruitflies and tracked 32 distinct points to describe the pose of the head, body,wings and legs, with an error rate of <3% of body length. We recapitulatedreported findings on insect gait dynamics and demonstrated LEAP's applicabilityfor unsupervised behavioral classification. Finally, we extended the method tomore challenging imaging situations and videos of freely moving mice.

参考文献：

Pereira et al (2019). Fast animalpose estimation using deep neural networks. Nat Methods. 2019Jan;16(1):117-125.

Nat Genet—全基因组分析和人工智能：全基因组深度学习分析发现非编码序列突变也与自闭症发生相关

英文摘要：

We address the challenge ofdetecting the contribution of noncoding mutations to disease with adeep-learning-based framework that predicts the specific regulatory effects andthe deleterious impact of genetic variants. Applying this framework to 1,790autism spectrum disorder (ASD) simplex families reveals a role in disease fornoncoding mutations-ASD probands harbor both transcriptional- and post-transcriptional-regulation-disruptingde novo mutations of significantly higher functional impact than those inunaffected siblings. Further analysis suggests involvement of noncodingmutations in synaptic transmission and neuronal development and, taken togetherwith previous studies, reveals a convergent genetic landscape of coding andnoncoding mutations in ASD. We demonstrate that sequences carrying prioritizedmutations identified in probands possess allele-specific regulatory activity,and we highlight a link between noncoding mutations and heterogeneity in the IQof ASD probands. Our predictive genomics framework illuminates the role ofnoncoding mutations in ASD and prioritizes mutations with high impact forfurther study, and is broadly applicable to complex human diseases.

参考文献：

Zhou et al (2019). Whole-genomedeep-learning analysis identifies contribution of noncoding mutations to autismrisk. Nat Genet. 2019 Jun;51(6):973-980.

NatMethods—质谱和人工智能：SIRIUS 4是一个可快速将质谱图转变为代谢物结构信息图的工具

英文摘要：

Massspectrometry is a predominant experimental technique in metabolomics andrelated fields, but metabolite structural elucidation remains highlychallenging. We report SIRIUS 4 (https://bio.informatik.uni-jena.de/sirius/),which provides a fast computational approach for molecular structureidentification. SIRIUS 4 integrates CSI:FingerID for searching in molecularstructure databases. Using SIRIUS 4, we achieved identification rates of morethan 70% on challenging metabolomics datasets.

参考文献：

Dührkopet al (2019). SIRIUS 4: a rapid tool for turning tandem mass spectra intometabolite structure information. Nat Methods. 2019 Apr;16(4):299-302.

NatBiotechnol—单细胞测序和人工智能：科学家开发Scanorama算法可高效的整合来源不同的单细胞测序数据

英文摘要：

Integrationof single-cell RNA sequencing (scRNA-seq) data from multiple experiments,laboratories and technologies can uncover biological insights, but currentmethods for scRNA-seq data integration are limited by a requirement fordatasets to derive from functionally similar cells. We present Scanorama, analgorithm that identifies and merges the shared cell types among all pairs ofdatasets and accurately integrates heterogeneous collections of scRNA-seq data.We applied Scanorama to integrate and remove batch effects across 105,476 cellsfrom 26 diverse scRNA-seq experiments representing 9 different technologies.Scanorama is sensitive to subtle temporal changes within the same cell lineage,successfully integrating functionally similar cells across time series data ofCD14⁺ monocytes at different stages of differentiation intomacrophages. Finally, we show that Scanorama is orders of magnitude faster thanexisting techniques and can integrate a collection of 1,095,538 cells in just~9 h.

参考文献：

Hie etal (2019).  Efficient integration ofheterogeneous single-cell transcriptomes using Scanorama.

NatBiotechnol. 2019 Jun;37(6):685-691.

NatBiotechnol—蛋白质测序和人工智能：深度神经网络帮助科学家快速发现蛋白质氨基酸序列中的信号肽

英文摘要：

Signalpeptides (SPs) are short amino acid sequences in the amino terminus of manynewly synthesized proteins that target proteins into, or across, membranes.Bioinformatic tools can predict SPs from amino acid sequences, but most cannotdistinguish between various types of signal peptides. We present a deep neuralnetwork-based approach that improves SP prediction across all domains of lifeand distinguishes between three types of prokaryotic SPs.

参考文献：

Almagroet al (2019). SignalP 5.0 improves signal peptide predictions using deep neuralnetworks. Nat Biotechnol. 2019 Apr;37(4):420-423.

Cell—mRNA剪切和人工智能：深度学习可从Pre-mRNA原序列中预测剪切的连接点

英文摘要：

Thesplicing of pre-mRNAs into mature transcripts is remarkable for its precision,but the mechanisms by which the cellular machinery achieves such specificityare incompletely understood. Here, we describe a deep neural network thataccurately predicts splice junctions from an arbitrary pre-mRNA transcriptsequence, enabling precise prediction of noncoding genetic variants that causecryptic splicing. Synonymous and intronic mutations with predictedsplice-altering consequence validate at a high rate on RNA-seq and are stronglydeleterious in the human population. De novo mutations with predictedsplice-altering consequence are significantly enriched in patients with autismand intellectual disability compared to healthy controls and validate againstRNA-seq in 21 out of 28 of these patients. We estimate that 9%-11% of pathogenicmutations in patients with rare genetic disorders are caused by this previouslyunderappreciated class of disease variation.

参考文献：

Jaganathanet al (2019). Predicting Splicing from Primary Sequence with Deep Learning. Cell.2019 Jan 24;176(3):535-548.e24.

NatMethods—冷冻电镜和人工智能：深度学习可从冷冻电镜图像中快速解析出蛋白质二级结构

英文摘要：

Althoughstructures determined at near-atomic resolution are now routinely reported bycryo-electron microscopy (cryo-EM), many density maps are determined at anintermediate resolution, and extracting structure information from these mapsis still a challenge. We report a computational method, Emap2sec, thatidentifies the secondary structures of proteins (α-helices, β-sheets and otherstructures) in EM maps at resolutions of between 5 and 10 Å. Emap2sec uses athree-dimensional deep convolutional neural network to assign secondarystructure to each grid point in an EM map. We tested Emap2sec on EM mapssimulated from 34 structures at resolutions of 6.0 and 10.0 Å, as well as on 43maps determined experimentally at resolutions of between 5.0 and 9.5 Å.Emap2sec was able to clearly identify the secondary structures in many mapstested, and showed substantially better performance than existing methods.

参考文献：

Maddhuriet al (2019). Protein secondary structure detection in intermediate-resolutioncryo-EM maps using deep learning. Nat Methods. 2019 Sep;16(9):911-917.

2019年十大研究进展名录

1. 年终盘点：2019年帕金森病十大基础研究进展

2. 年终盘点：2019年帕金森病十大临床研究进展

3. 年终盘点：2019年阿尔茨海默病十大基础研究进展

4. 年终盘点：2019年阿尔茨海默病十大临床研究进展

5. 年终盘点：2019年神经科学领域十大基础研究进展

6. 年终盘点：2019年抑郁症领域十大基础研究进展（一半来自中国）

7. 年终盘点：2019年脑血管病领域十大基础研究进展

8. 年终盘点：2019年神经炎症领域十大基础研究进展

9. 年终盘点：2019年神经活动记录十大基础研究进展

2018年十大研究进展名录

1.盘点2018年阿尔茨海默病十大研究突破

2.盘点2018年帕金森病十大研究突破

3. 盘点2018年神经科学二十大研究突破

4. 盘点2018年渐冻症（ALS）十大研究进展

5. 盘点2018年全球脑卒中十大研究进展

6. 盘点2018年神经影像十大研究进展

7. 盘点2018年神经炎症领域的十大研究突破

8. 盘点2018年神经变性痴呆十大研究突破

9. 2018年神经科学“学习和记忆”领域十大研究进展

10. 2018年抑郁症领域的十大研究突破

11. 2018年痛觉和疼痛领域的十大研究突破

12. 2018年的神经干细胞研究十大研究进展

13. 2018年的神经干细胞研究十大研究进展

14. 2018年的十大睡眠研究突破

15. 2018年“衰老和长生不老”领域的十大研究突破

16. 2018年自闭症领域的十大研究突破

欢迎加入60个“神经科学临床和基础社群”

1、神经科学临床和基础主群（500人）已满；

2、神经科学临床和基础Alzheimer亚群；

3、神经科学临床和基础Parkinson亚群；

4、神经科学临床和基础cerebrovascular亚群；

5、神经科学临床和基础Depression亚群；

6、神经科学临床和基础Movement disorders亚群；

7、神经科学临床和基础Neuroimmunology亚群；

8、神经科学临床和基础Psychiatry亚群；

9、神经科学临床和基础Neuroimaging亚群；

10、神经科学临床和基础Neurogenetics亚群；

11、神经科学临床和基础Neurodegeneration亚群；

12、神经科学临床和基础Epilepsy亚群；

13、神经科学临床和基础Sleep亚群；

14、神经科学临床和基础Neural Development亚群；

15、神经科学临床和基础Electrophysiology亚群；

16、神经科学临床和基础Neural circuits亚群；

17、神经科学临床和基础神经调控和脑机接口亚群；

18、神经科学临床和基础人工智能亚群；

19、神经科学临床和基础重大疾病和疑难病亚群；

20、神经科学临床和基础衰老和永生亚群；

21、神经科学临床和基础周围神经病群；

22、神经科学临床和基础神经肌肉疾病群；

23、神经科学临床和基础视觉系统研究群；

24、神经科学临床和基础疼痛研究群；

25、神经科学临床和基础Emotion研究群；

26、神经科学临床和基础意识研究群；

27、神经科学临床和基础Learning & Memory亚群；

28、神经科学国自然基金申请交流群；

29、神经科学ALS/FTD交流群；

30、神经科学脑外伤和脊髓外伤研究群；

31、神经科学儿科神经病学交流群；

32、神经科学Autism & ADHD研究群；

33、神经科学大数据和组学研究群；

34、神经科学非编码RNA研究群；

35、神经科学schizophrenia研究群；

36、神经科学Non-human primate研究群；

37、神经科学神经损伤与修复研究群；

38、神经科学Epigenetics研究群；

39、神经科学神经介入和静脉溶栓亚群；

40、神经科学计算神经科学亚群；

41、神经科学基因治疗交流群；

42、神经科学细胞治疗交流群；

43、神经科学纳米药物治疗交流群；

44、神经科学中医药治疗交流群；

45、神经科学免疫调节治疗交流群；

46、神经科学类器官和类脑研究交流群；

47、神经科学语言研究交流群；

48、神经科学深度学习和神经网络交流群；

49、神经科学类神经元和类脑器件设计交流群；

50、神经科学半人半机器人交流群；

51、神经科学感染性疾病研究群；

52、神经科学神经系统肿瘤研究群；

53、神经科学星型和小胶质细胞研究群；

54、神经科学神经外科研究群;

55、神经科学系统论和复杂性研究交流群；

56、神经科学脑肠轴和Microbiota交流群；

57、神经科学虚拟现实、增强现实和混合现实交流群；

58、神经科学临床试验和流行病学研究交流群；

59、神经科学单细胞测序研究交流群；

60、神经科学蛋白质解析交流群。

如果想入群，请加我微信（qingyierjing），并回复要加入的群，我会将您拉入群中。

20个神经科学领域的突破可能获得诺贝尔奖

1. 意识研究：意识的本质、组成、运行机制及其物质载体；不同意识层次的操控和干预，意识障碍性疾病的治疗。

2. 学习和记忆的机制及其调控：记忆的形成和消退机制，记忆的人为移植和记忆的人为消除等；

3. 痴呆研究：阿尔茨海默病的机制和治疗研究，血管性痴呆、额颞叶痴呆、路易体痴呆的机制研究和治疗。

4. 睡眠和睡眠障碍的机制和干预研究。

5. 情绪研究：喜、怒、哀、恐等基本情绪的机制和相关疾病的治疗。

6. 计算和逻辑推理的神经科学基础研究。

7. 语言的神经科学基础研究。

8. 视觉图像形成和运用的神经科学基础研究。

9. 创造力、想象力和艺术文学创造的神经基础研究。

10. 痛觉的神经科学基础及其干预研究

11. 性行为研究：性行为的神经科学基础研究和性行为的调控和干预。

12. 脑和脊髓损伤的机制及其干预研究，包括脑卒中、脊髓损伤机制研究，神经干细胞移植研究，新型神经修复技术，神经康复技术。

13. 精神类疾病的机制和干预研究：自闭症、精分、抑郁症、智能障碍、药物成瘾等；

14. 运动神经元病等神经变性病机制研究及其干预。

15. 衰老的机制和永生研究，包括大脑衰老的机制和寿命延长研究。

16. 神经系统遗传病的机制研究及基因治疗。

17. 神经操纵和调控技术：光遗传技术、药物遗传技术、基因编辑技术、经颅磁刺激、深部脑刺激和电刺激等。

18. 脑组织兼容性电子微芯片及脑机互动装置研究，包括脑机接口、神经刺激芯片、记忆存储芯片，意识存储芯片，人脑非语言互动装置等。

19. 半人半机器人的设计、完善和修复技术：包括任何机械肢体的人类移植，大脑移植入机器体内等。

20. 新型大脑成像和神经元活动记录技术：高分辨率成像技术、大型电极微阵列技术等。