AI and Life Sciences

AI And life sciences

Machine learning helps mathematicians discover two conjectures , And using machine learning 、 The combination of multi-scale modeling and high-performance computing solves the problem of real-time simulation of very large-scale quantum random circuits , Let people see the application of artificial intelligence in scientific research , Processing data 、 Great potential in designing new experiments and creating more efficient computational models .

AI for Science Emerging research areas appear , It is expected to bring about changes in the paradigm of scientific research ！ On the rise AI for Science It is expected to promote the deep integration of data-driven and theoretical deduction .

In the field of computer ,Jeff Dean It is also emphasized in disease diagnosis , Especially in the application of medical imaging , For example, in improving breast cancer screening 、 Detect lung cancer 、 Accelerate cancer radiotherapy 、 Mark exception X X-ray and prostate cancer biopsy .

Another noteworthy direction is to use NLP Technology to analyze structured data and medical records , Assist clinicians to provide more accurate diagnosis and care .

AI By discovering hidden patterns in massive data , Assist mathematicians to put forward two conjectures , Let people see AI More potential in scientific research ,AI for Science Become a new hot word .

Baidu Research Institute believes ,AI It is expected to bring about changes in the paradigm of scientific research , The integration of data-driven and theoretical deduction will play a role in more disciplines .

The drug development process is a costly thing , Whether in terms of time or money invested . Even developing a single component requires the joint efforts of hundreds of researchers . One of the most interesting applications of modern AI is drug discovery . Researchers predict , The most advanced AI algorithms can be used to speed up the whole process . for example , Cell counting is an active research field in Biology . Vision AI Systems and computer vision can help complete it with unimaginable accuracy in a few seconds .

AI Let gene editing find targets more accurately and quickly , AI Help to make a significant breakthrough in protein structure prediction . The deep integration of the two will significantly shorten the drug R & D cycle 、 Reduce R & D costs , Promote precision medicine and personalized diagnosis and treatment .

Life science ushers in the data-driven era , The core representative is AlphaFold2.

AlphaFold2 And the open source of human proteome data sets has detonated computational biology again ！

Artificial intelligence predicts protein structure

12 month 15 Japan ,Nature Released 《2021 Top ten science news of the year 》;12 month 17 Japan ,Science Followed by , released 《2021 Top ten scientific breakthroughs of the year 》.Nature and Science All will 「 Artificial intelligence predicts protein structure 」 Rated as the most important discovery of the year ,Science It is listed as “2021 Ten scientific breakthroughs in ” first .

For a long time , The prediction of protein structure has always been a hotspot and difficulty in the field of Biology . There are three traditional methods of protein structure detection ：X Ray crystallography 、 Nuclear magnetic resonance and cryoelectron microscopy . But these methods cost more , The research cycle is long , And the progress is limited .

Artificial intelligence has pressed the fast forward key for this difficult problem that has plagued the biological community for decades .

This year, 7 month , There are two major protein structures AI Prediction algorithm —— DeepMind Of AphaFold2 And the University of Washington and other institutions RoseTTAFold Open source one after another .

AphaFold2“ Unlock ”98% Human proteome

7 month 16 Japan ,DeepMind stay Nature publish one’s thesis , Declared utilized Alpha Fold2 Predicted 35 Ten thousand protein structures , covers 98.5% Human proteome , And others 20 An almost complete proteome of an organism . The research team also announced AlphaFold2 Open source code and technical details .

RoseTTAFold The protein structure can be calculated in ten minutes

On the same day , Institute of protein design, University of Washington David Baker Professor research group and other cooperative institutions are in Science Last published , Announced its open source protein prediction tool RoseTTAFold The results . The research team explored the network architecture combining relevant ideas , And get the best performance through the three track network . The structure prediction accuracy produced by the three track network is close to CASP14 Medium DeepMind Team AlphaFold2, And faster 、 Lower computer processing power required . Use only one game computer , The protein structure can be reliably calculated in just ten minutes .

DeepMind Open source AlphaFold2, Predict 98.5% Human protein structure . meanwhile ,AlphaFold 2 Selected last year Science Top ten breakthroughs of the year , It's called structural biology “ revolutionary character ” Breakthrough 、 A milestone in protein research .

2018 Year of AlphaFold The neural networks used are similar ResNet Residual convolution network , here we are AlphaFold2 It draws Transformer framework .

AlphaFold2 Using multiple sequence alignment , The protein structure and biological information are integrated into the deep learning algorithm . The appearance of it , It can better predict the probability of protein binding to molecules , So as to greatly accelerate the efficiency of new drug research and development .

But after further development , Data bottlenecks cannot be ignored ： Insufficient high-quality R & D data , And the available data of pharmaceutical research and development is inversely proportional to the target value . However, there are already corresponding solutions in the industry , For example, establish a drug big data laboratory 、 Multidisciplinary integration and other methods .