《AlphaSense：2024人工智能如何重塑藥物研發研究報告：進展及挑戰（英文版）（16頁）.pdf》由會員分享，可在線閱讀，更多相關《AlphaSense：2024人工智能如何重塑藥物研發研究報告：進展及挑戰（英文版）（16頁）.pdf（16頁珍藏版）》請在三個皮匠報告上搜索。

1、How Artificial Intelligence isReshaping Drug DevelopmentExecutive SummaryIn the challenging landscape of drug development,where bringing a medication to market hasgotten slower and costlier over time,artificial intelligence(AI)solutions are gaining traction.Thisissue,coupled with escalating public a

2、nd political calls for more affordable medications,hasprompted a surge in AI-based initiatives aimed at reducing costs and expediting developmenttimelines.Additional momentum behind AI adoption in drug development is being fueled byadvancements in AI technology,heightened awareness,and improved comp

3、uter processingspeed.Estimates point to a substantial total addressable market of around$50 billion for AI-enableddrug development,with the expectation that 30%of new drugs will be discovered using AI by2025.Notably,the field is witnessing a proliferation of players,categorized into biopharmacompani

4、es with a robust AI focus and providers of AI tools working in tandem with largebiopharma players.Meanwhile,collaboration is on the rise with big tech companies like Nvidia,Alphabet and Amazon.In clinical trials,AI-derived drug candidates are gaining traction,while future AI applications areexpected

5、 to expand into challenging areas such as biologics,oncology,and rare diseases.Experts are enthusiastic about the big picture while acknowledging that AI-based drugdevelopment is still in an early and uncertain stage,and offered varied opinions on its progressand challenges.Lets explore this topic a

6、nd the key industry debates by leveraging AlphaSense,including first-person insights from our expert transcript library.A Ripe Time for AI SolutionsDrug development is an exceedingly time-consuming and expensive process.On average,ittakes more than 12 years and$1.3 billion to bring just one drug to

7、market.Factor in the highrate of drug candidates that never make it to market at all,mainly because of clinical trialfailures(roughly 90%)and the average cost per approved drug balloons to$2.6 billion.Against this backdrop,public and political demands for more affordable medications aremounting.Its

8、no wonder drug makers are racing to find AI-based solutions aimed at reducingdevelopment costs and shortening lead times.Also fueling the trend are AI technology advances and growing AI awareness,according toAlphaSense experts,as well as improved computer processing speed.Another driver is therecent

9、 explosion in the size and volume of data sets that feed AI models.AI models areaggregating data collected from diverse sources such as electronic health records,clinical trials,genomic sequencing,prescriptions,insurance claims,and remote monitoring devices likefitness trackers,pacemakers and glucos

10、e monitors.As the cost of clinical trials continues to go out of sight,and there is pressure on drugs tocome down in cost,this AI drug discovery stuff will just continue.It makes sense;youhave all this data,why not analyze it versus trying to run an experiment yourself?”“Using computer-driven algori

11、thms to enhance drug design is not new.This has beengoing on for decades.Whats different now?the emergence of ChatGPT and things likethis that are now accessible to almost anybody with a computer.This,I think,captured theimagination of a much broader swath of people.On the back of that is the fact t

12、hat thecomputer processors are much faster.”Former Senior Vice President,Biotech Company|Expert TranscriptDrug companies enthusiasm for AI-based solutions is reflected in the AlphaSense researchplatform,where documents mentioning AI in drug development have surged during the past 12months.Source:Alp

13、haSenseDrug Development 101Drug development is the process by which new medications are identified,designed,anddeveloped for treating or preventing diseases.Heres a quick overview of how drugdevelopment works:Source:The University of TexasHowever,the traditional drug development process has several

14、drawbacks,which we explore ingreater detail in the following section.These drawbacks have contributed to a confoundingphenomenon whereby drug development has gotten slower and costlier over time despitetechnological improvements and increased investment in research and development(R&D).According to

15、Erooms Law(Moores Law spelled backwards),the number of new drugsapproved per$1B spent on R&D has halved roughly every nine years since 1950.Erooms Law:New Drugs Per$1B in R&D Spend Halves Every 9 YearsSource:Nature Reviews Drug DiscoveryChallenges with Traditional Methods Equate toAI OpportunitiesAI

16、 refers to the simulation of human intelligence in machines that are programmed to think,learn,and problem-solve like humans.The goal of AI is to develop systems that can perform,enhance,and augment tasks that typically require human intelligence,such as visualperception,speech recognition,decision-

17、making,and language translation.Drug development is attracting a disproportionate share of AI investment dollars flowing into thehealthcare sector because of its compelling return on investment.Estimates are scarce,but aBCG report released last year said that AI could yield time and cost savings of

18、at least25%50%in the drug discovery steps leading up to the preclinical stage.Drug discovery is where the biggest investment is being made at the moment Indiscovery,anything that can drive improvement could have dramatic consequences onthe return of investment downstream,the net present value,for ex

19、ample.Its more or lessunderstandable that the most AI investment is there.”Associate Director,Pharmaceutical Company|Expert TranscriptIn comparison to traditional methods,AI-enabled drug development promises to expand thenumber of potential therapeutic starting points,while identifying failures earl

20、ier in the researchprocess and accelerating the pace of bringing drugs to market.As a grim reaper,AI is really good.It can catch,lets say,possible toxicology effects sothat you can kill a potential molecule much earlier.Does it increase the probability oftechnical success?No.What it does is it reduc

21、es the aggregate cost.You will still need 22projects to get one candidate,but you will be able to kill the first 21 much earlier.”Associate Director,Pharmaceutical Company|Expert TranscriptAI-Enabled vs.Traditional Drug DevelopmentSource:Recursion PharmaceuticalsBelow,we explore some of the specific

22、 issues AI addresses in the drug development process.Identifying Drug Targets and CandidatesIssue:Traditional methods are mostly based on singular hypothesis-driven approaches and arelimited by human capacity to analyze multi-dimensional data and prone to human error andbias.Solution:AI takes a more

23、 holistic approach by analyzing vast and diverse datasets to expediteand improve the accuracy of identifying drug targets and candidates.AI and ML and analytics,basically these are ways to really take vast amounts of disparatedata,and when queried in the correct way,can provide some insights that yo

24、u just cant dosingle-handedly in the human mind.”Former Managing Director,Biotech Company|Expert TranscriptPreclinical TestingIssue:Predicting the safety and efficacy of drug candidates in preclinical testing is a significantchallenge and dependent on trial-and-error in the lab.Solution:AI uses pred

25、ictive tools that can help reduce experimental failures and animal testingrequirements.Nobody wants to kill millions of animals through high-throughput in vivo experiments.There is a huge market opportunity for AI there.”Associate Director,Pharmaceutical Company|Expert TranscriptClinical Trial Desig

26、n and Patient RecruitmentIssue:Designing efficient clinical trials with suitable patients can be challenging.Clinical trialsoften involve relatively small and homogeneous patient populations.Solution:AI assists in optimizing trial design,predicting patient responses,and improvingpatient recruitment.

27、It also analyzes patient data from real-world sources across a widerdemographic,which has become increasingly important for regulatory approval according toFDA guidance issued last year.A key opportunity for AI is the optimization of clinical trials,optimization of clinical trialdesign,optimization

28、of the numbers of subjects,optimization of the patient selection,patient selection criteria.Thats a huge area of opportunity.”Former Senior Vice President,Biotech Company|Expert TranscriptPersonalized MedicineIssue:Identifying patient subgroups that respond differently to treatments is critical forp

29、ersonalized medicine.Solution:AI analyzes patient data,including genetic information,to identify biomarkers andstratify patients based on their likelihood of responding to specific treatments.Natural Language Processing(NLP)for Literature MiningIssue:Keeping up with the vast and constantly evolving

30、scientific literature is difficult.Solution:AI,specifically NLP,helps researchers extract relevant information from literature,patents,and clinical trial reports to aid decision-making.Drug RepurposingIssue:Finding new uses for existing drugs is challenging.Solution:AI can analyze large datasets to

31、identify potential connections between drugs anddiseases,allowing drugs to be repurposed.Market Opportunity Attracting Plethora ofPlayersEstimates of the total addressable market(TAM)for AI-enabled drug development are scarce,though TD Cowen and Morgan Stanley both approximate the TAM at$50B.An esti

32、mated 30%of new drugs are expected to be discovered using AI technology by 2025,up from zero today,and AI is expected to lead to 50 novel therapies over a 10-year period.AI Drug Development UniverseSource:TD CowenThe field of players looking to capitalize on AI drug development trends is plentiful a

33、nd falls intotwo main categories:biopharma companies with a strong AI focus and providers of AI tools.Biopharma Companies with a Strong AI FocusThese companies use internal databases,algorithms,and computing power to discover drugcandidates and move them into clinical trials.In addition,this group o

34、ften depends on externalpartners to provide AI tools or applications.Merck and some of these larger companies are well-positioned to use treasure troves oftheir own internal data repositories,high-quality data to fish out certain patterns in thatdata across whether its chemistry or pharmacokinetics

35、or safety talks.”Senior Scientist,AI Tools Provider|Expert TranscriptProviders of AI ToolsThese companies serve as partners and collaborators with biopharma players by offering vastdatabases for feeding AI models,deep AI expertise,or AI platforms that can be applied across arange of products and the

36、rapeutic areas.Collaborations on the Rise Among Big Pharma and AI Tools ProvidersSource:TD CowenExperts believe AI tools providers that have the most established relationships with largebiopharma companies are best positioned in the market.A number of companies that have started out like Recursion,E

37、xscientia,BenevolentAI,Owkin,maybe theyre five-,six-,seven-years old,theyve really started to come into theirown now and get the credibility with the pharma companies that theyre trying to partnerwith.Theyre actually delivering now in terms of the promise of the application of advancedanalytics.”“Wh

38、at Im just starting to see is that people are leaving the pharma companies and joiningthese AI tools provider companies.Its been really important to get experienced drugdevelopers in these AI analytics companies,and so thats happening now.”Former Managing Director,Biotech Company|Expert TranscriptOn

39、e expert said providers of AI tools that offer a combination of proprietary data and AIcapabilities are particularly well-positioned:You have a category of companies that are hybrid.They are generating their own dataand theyre also having AI/ML capabilities.Now,these are the most interesting onescom

40、panies like Recursion,companies like Insitro who have understood that the biggestdifferentiator in the world of AI is not really the technology itself,but rather the data.There are not that many of those companies,but you see that the market is moving inthis direction.”Associate Director,Pharmaceuti

41、cal Company|Expert TranscriptOther experts said no clear winners among the AI tools providers have distinguishedthemselves:Its a Wild West right now In my opinion,no ones emerged.There are definitely somecompanies that are more notable than others.Some have received large investmentrounds and things

42、 of that nature that help to raise their profile.Im not really convinced yettheres a single company or even type of company thats really emerged as the go-tosource or a go-to source.”Former Senior Vice President,Biotech Company|Expert TranscriptMeanwhile,big tech players such as Nvidia,Alphabet,Amaz

43、on,and Microsoft are expandinginto biopharma to provide the underlying computing power and tools to enable AI drugdevelopment at scale.Clinical Pipeline Reflects AI MomentumAs of last year,there were 15 AI-derived drug candidates in clinical trials,and the pipeline ofAI-derived assets continues to e

44、xpand.One company in particular has been garneringattentionHong Kong-based Insilico Medicinefor making the first drug entirely based on AI.Insilico Medicine developed INSO18-055 for the treatment of idiopathic pulmonary fibrosis(IPF)with the first AI-discovered molecule that was based on a novel AI-

45、discovered target.Insilico Medicine has been democratizing their own internal platform,which I think isreally cool.Theyre essentially saying,Weve developed these algorithms,the generativealgorithm to identify new molecules and also our predictive models,and were makingthose models and algorithms ava

46、ilable to you as a fee-for-service.Theyre using thatinternally for a set of targets and then also partnering with large-and medium-sizedpharma companies.”Senior Scientist,AI Tools Provider|Expert TranscriptNotably,Insilicos drug was selected as a preclinical candidate in February 2021less than 18mon

47、ths after the project beganand announced its first Phase I trial just nine months later.Altogether,it took INSO18-055 under 30 months to progress to Phase 1about half the time ittakes in traditional drug discovery.In doing so,INSO18-055 was the first fully derived AI drug tobe tested in human trials

48、.INSO18-055 has since entered Phase II clinical trials,and,ifsuccessful,the drug will proceed to further studies with larger cohorts,potentially leading toPhase III trials.Focus Increasing in Challenging AreasAI applications in drug development are expected to expand into areas posing particularlyco

49、mplex problems,including biologics,oncology,and rare diseases.BiologicsSo far,the majority of AI-based drug development has been focused on small-molecule drugs,which are chemical-based,have simple structures,and make up the majority of thepharmaceuticals market(90%of global pharmaceutical sales in

50、2021).But in recent years,drug development has been moving toward large-molecule drugs,or biologics,which arederived from living organisms such as proteins or antibodies.Biologics offer greater targetability and less toxicity than small molecules,and have facilitatedpersonalized medicines such as CA

51、R-T cell therapy for cancer.Yet biologics have complexstructures,which makes developing them both difficult and expensive.On average,the dailydose of a biologic costs roughly 22 times more than that of a small molecule.AI lends itself to the complexities of biologics because it is able to rapidly an

52、alyze large-scaledata sets,predict protein structures,and design new molecules.Consequently,there is agrowing list of AI-derived biologics projects in the drug development pipeline.There is tremendously promising work in generating protein sequences for biologics thatwill hopefully start delivering

53、products within maybe the next five years or so.”Former Vice President,AI Tools Provider|Expert TranscriptFurthermore,there are now 80 or so AI-focused companies working in the biologics arena,mostof which were founded within the past seven years.Companies Working in AI-Driven Development of Biologi

54、csSource:NatureOncologyIn terms of therapeutic areas,AI is expected to continue to focus its powerful capabilities ondifficult-to-treat disease types,with oncology at the top of the list.Oncology is attractive not onlybecause of its complexity,but also because of the prevalence of cancer patients an

55、d,in turn,patient data to feed AI models.Indications of Focus Across Active AI-Enabled Clinical TrialsSource:TD CowenRare DiseasesOn the other hand,rare diseases do not have large commercial markets,yet AI-based drugdevelopment is also trending toward rare disease.This is explained by regulatory inc

56、entives,such as orphan drug designations,which have encouraged pharmaceutical companies to investin treatments for rare diseases.But AIs ability to overcome unique challenges in rare diseasesis also a factor.These challenges include a dearth of patient data as well as experts with sufficient knowled

57、geand experience to run trials,both owing to the fact that rare disease patient populations aresmall and variable.In the United States,rare disease is defined as a disease or condition thataffects less than 200,000 people.In addition,rare diseases most commonly develop in early childhood.Conducting

58、clinical trialsin children is complicated not only due to ethical questions but also due to variability in childrensphysiology and the ways in which drugs act in their bodies.AI can help bridge these gaps by mining existing drug databases for candidates,which can thenbe repurposed for rare diseases.

59、Repurposing this data also has the benefit of reducing thenumber of preclinical and clinical safety studies required because the FDA has alreadyconsidered safety effects in the approved drug.In addition,AI can help identify subpopulationswithin a group of rare disease patients,allowing for more targ

60、eted and efficient clinical trials.Looking AheadExperts agree that although AI-based drug development is exciting,we are still in early innings,and chaotic innings at that:Theres a lot of confusion in the industry about how AI is being applied.”Senior Scientist,AI Tools Provider|Expert TranscriptLik

61、ewise,visibility is limited into how the field will advance and how quickly,with experts offeringa range of opinions.The same expert quoted above offered a relatively upbeat view:In two years Im cautiously optimistic that well be able to start seeing the reward thatAI offers the industry,but you jus

62、t dont know.I hate to be ambiguous like that,but thatsmy take.”Senior Scientist,AI Tools Provider|Expert TranscriptAnother expert was more reserved:At the moment I see a lot of reasonable expectations,but there are also people expectingthings too soon in some areas.There are areas where people hope

63、AI will work,but itmay not work at all.In cheminformatics,in model analysis for small molecule discovery,inbinding prediction,in generation of biologics,Im pretty sure it will work pretty well.Inthings like target discovery using cell biological models,Im less sure.”Former Vice President,AI Tools Pr

64、ovider|Expert TranscriptAs dynamics continue to unfold in AI-based drug development,new questions will most certainlyarise and AlphaSense will be there to monitor developments with the help of first-person expertinsights.Stay tuned!Ready to dive deep into the expert transcript library?Start your free trial of AlphaSensetoday.