數位化｜AI 如何改變醫藥的研究模式，百靈佳殷格翰運用人工智慧造福患者 Artificial Intelligence serving patients
百靈佳殷格翰將人工智慧運用於整個價值鏈中，從早期研究、開發、生產到產品分銷。具體的目標因實際應用情況而定，包含為患者提供更優質、更快速取得或者更安全的產品。詳情請參閱以下聊天內容。
為什麼大家都在討論 AI 人工智慧？AI 人工智慧已經存在多年，並且已經融入工業和日常生活的應用中。2022 年時，ChatGPT 和其他類似應用程式，讓人工智慧—更明確的說，是生成式人工智慧—更廣泛地為大眾使用。生成式人工智慧讓人既驚奇又害怕，成為媒體關注的焦點，也成為受人矚目的投資標的。到目前為止，AI 主要用於協助一般的網路使用者產出文字或圖像等內容。
AI 的商業應用是否被過度誇大了？AI 的商業應用並未被過度誇大。雖然目前只有相對少數的企業依賴 AI 進行產出，因此某些觀察者認為AI被過度炒作。但 AI 確實對越來越多企業帶來影響。例如，多年來， AI 在百靈佳殷格翰發揮著積極且不斷擴展的角色，員工們正廣泛運用生成式 AI 來加速研究。
百靈佳殷格翰用哪些方式將 AI 與業務融合？百靈佳殷格翰將人工智慧運用於整個價值鏈中，從早期研究、開發、生產到產品分銷。例如，銷售和行銷部門運用 AI 優化可能對某些產品有興趣的醫療專業人員的推廣。其他 AI 應用案例包括：提高生產線效率、確認符合監管程序與要求，提供患者創新的醫療方案。最近，百靈佳殷格翰率先採用生成式 AI 推動 iQNow 知識平台，幫助百靈佳殷格翰的科學家們更能在大量的研究文獻中準確提取知識與資訊。自 2023 年初推出以來的9個月內，百靈佳殷格翰 iQNow 與大型語言模型（Large Language Model，LLM）的互動功能結合， 增強的功能讓不重複使用者（Unique User）由 1,000 人增加為 25,000 人。
百靈佳殷格翰如何成為生成式 AI 的先驅？百靈佳殷格翰做好充足的準備。多年前，公司收到多個部門反應他們被各種資訊淹沒，各資料庫都用不同的方式編輯，文件也分散在各處；收到回饋的時候百靈佳殷格翰就已經在尋找解決方法，希望能改善文件存取方式。公司看到 AI 帶來的進步，將其視為成長改變的契機。 IQNow 知識平台的主要功能是運用複雜的演算法檢索特定資訊，並且在生成式 AI 的大語言模型出現後，進一步強化了既有功能。
將公司的專有數據上傳到雲端 AI 應用程式是否有風險？為確保專有數據的完全安全， iQNow 知識平台連結到微軟（Microsoft）的 Azure Open AI 服務，這是一種安全地將這些互補技術結合在一起的方法。公共數據源和專有數據源之間保持完全分離。
iQNow 知識平台如何改變百靈佳殷格翰的研究發展？百靈佳殷格翰的研究人員能使用 iQNow 知識平台檢視約 7 億個文件資料庫，快速辨識特定研究領域的專家。研究人員可以向它提問，它會像身邊的同事一樣，給出可靠的答案。透過遵循人類對話的邏輯， iQNow 知識平台 幫助他們分析所遇到的任何研究問題。研究人員可以對 iQNow 知識平台進一步提出更深層次的問題，或請它提供可參考的文件資料。
生成式 AI 對百靈佳殷格翰的研究有明顯的幫助嗎？答案是肯定的。生成式 AI 對百靈佳殷格翰的研究提供了極大的幫助。 iQNow 知識平台推出後，有一群特定的團隊運用它強化研究內容。當 ChatGPT 出現之後，生成式 AI 成為更實用的日常工具；百靈佳殷格翰隨即將 iQNow 知識平台與大語言模型的互動功能結合。系統上線 70 天內，節省了近 15 萬工時。在前 9 個月內幫助節省約 60 萬工時，相當於 6000 萬歐元的研究支出。
除了節省時間， AI 的應用對公司的研究有何效益？AI 不再僅是幫助百靈佳殷格翰的研究人員尋找特定資料或專家的工具，而是可以讓他們運用大量研究文獻資料庫中所取得的知識進行有序對話。這直接有助於加速百靈佳殷格翰的藥物研究和研發進程。
百靈佳殷格翰還用 AI 做了什麼？百靈佳殷格翰用 AI 做了很多不同的事情。 iQNow 知識平台為每個人提供生成式 AI ，現在這個功能也可透過 Copliot 等微軟應用程式達成。其他範例也展示了生成式 AI 在特定業務中的有效性：百靈佳殷格翰現在運用 AI 總結醫藥學術專員團隊取得的資訊。百靈佳殷格翰在該領域的第一個產品 KNERD 現已被研究人員廣泛使用。 ADAM 則是公司專門的分子設計高級助手，用以協助藥學化學家們尋找新的備選藥物。 |
這些案例展示了百靈佳殷格翰運用 AI 所產生的價值以及多樣性。與此同時，我們的 IT 部門也建立了一個強大的解決方案框架，為所有使用者提供標準的 AI 解決方案，提供可以開發並執行客製化方案的平台，大家都能找到符合各自業務和職位權責的方案建議。
未來 AI 將如何影響製藥業？
AI 預計將以多種方式改變製藥業。首先，將會加速藥物研發的速度，縮短識別可用以治療的化合物和標靶的過程。 AI 可以協助分子結構預測、從頭開始進行蛋白質設計，進而推動新藥的開發。
很快地， AI 也能支持低風險臨床試驗環境中的工作，加速試驗獲得監管批准的時間，更迅速地為有需要的患者提供新的治療方案。
除此之外， AI 也將加速個人化的藥物研發歷程 — 這最終將改善治療的結果。
為了探索量子電腦在藥物研發領域的潛力，百靈佳殷格翰在 2021 年時與 Google Quantum AI 合作。目前的進度如何？
訪問 Clemens Utschig-Utschig 百靈佳殷格翰 CTO 暨 IT 首席系統架構師（CTO & Chief Architect IT, Boehringer Ingelheim）
1. 目前利用量子電腦進行藥物研究方面的進展如何？
量子電腦或許可以實現高度加速的運算能力。化學領域可能會成為首個主要應用的領域。藥物設計時需要準確預測候選藥物如何與活細胞中的標靶相互作用。這必須要模擬成千上萬個原子在特定溫度下的表現，並且進行深入分析；傳統實驗無法做到這個地步。
量子電腦能夠執行高效率的化學計算，可以模擬系統的量子特性；傳統電腦無法負荷如此複雜的任務。目前的量子電腦仍處於早期開發的階段，僅能應用於非常小規模的系統。為了模擬與製藥相關的分子，我們需要更大、可靠的量子電腦，以及根據我們需求量身定制的量子算法。其中一個核心目標是實現與目前實驗相當的計算模擬環境，令實驗更易於運作。量子電腦的硬體正在迅速發展，每天都有新的量子演算法被開發出來，離實際應用已越來越近。
2. 百靈佳殷格翰目前學到了什麼？量子電腦什麼時候才能成為可以造福患者的工具？
為了探索量子電腦在藥物研發領域的潛力，百靈佳殷格翰在 2021 年時與 Google Quantum AI 合作。
我們一起探索了多種可以實際操作的方案，例如：找到量子算法來研究 P450 酶。 P450 酶在人體新陳代謝中扮演重要的角色，過去從未以量子計算進行分析。分析結果顯示，量子電腦的計算結果非常精確，明顯優於傳統的方式。
然而，即使採用了最適合的量子算法，也需要三天的計算時間，遠超過在工業設置中實際可行的範圍。我們目前正在努力開發新的算法，目標是將電腦運作的時間從數小時或數天縮短到幾分鐘。
目前我們也與多倫多大學合作進行研究，開發量子算法研究分子動力學，預測分子隨時間進行的運動方式。
我們的主要目標是預測候選藥物分子與標靶的結合程度；因此，我們開發了一種新的量子算法用於分子動力學，並在多個國際會議上展示我們的成果。
我們儘管在軟體、硬體和應用案例等方面取得穩定的進展，卻依然位於應用基礎研究的階段。
我們需要進一步的發展以充分發揮量子計算的經濟潛力。我們將持續努力推動發展，期望能在十年內提供與業界相關的應用案例。
3. 下一步是什麼？
目前要預測製藥業什麼時候能充分利用量子電腦的潛能，還言之過早。
硬體上必須要改進，也必須開發新的算法。除此之外，我們也要想出新的策略，才能在計算準確性跟計算時間之間找到折衷方案。
目前的主要重點在於減少量子算法的運作時間 — 畢竟量子計算的結果比傳統電腦的低精度計算更具優勢 — 同時探索新的應用方法。
換句話說，我們可以與合作夥伴一起積極貢獻我們的專業知識，面對各種挑戰。我相信未來幾年內將會有明確可見的進步。
我們的量子實驗室與合作夥伴的觀點已經發表並刊登在《Nature Physics》。
Artificial Intelligence serving patients
Boehringer Ingelheim is leveraging AI across its entire value chain, from early research through development, production to the distribution of products. The objectives vary depending on the specific application: better, more quickly available or safer products for patients. Find out more in the chat below.
Why are so many people talking about AI these days?AI has been around for many years already and is built into many applications in the industry and daily life. ChatGPT and other like-minded applications have made AI, and more specifically Generative AI, more widely available to the general public in 2022. The technology was thrust into the media spotlight as a growing object of wonder and fear, as well as a target for investment. So far, this has mainly allowed the average internet user to generate content such as text or images with the help of AI.
Isn't it possible the business applications for AI are being overhyped?The business applications for AI are certainly not all being overhyped. It's true that a relatively small number of businesses currently depend on AI to generate practical results, so it is unsurprising that some observers have argued that AI is being overly hyped. But the impact of AI at a growing number of companies is significant. For example, it has been playing an active and expanding role at Boehringer Ingelheim for years, with employees now widely using Generative AI to accelerate their research.
What are some of the ways that Boehringer has been incorporating AI into its business?Boehringer Ingelheim is using AI across the entire value chain, from early research through development, production and the distribution of products. For instance, the company's sales and marketing department uses AI to optimize the outreach to healthcare professionals who might be interested in learning about certain products. Other examples where AI is already used include solutions to improve the efficiency of the company's production lines, to ensure regulatory preparedness or to offer innovative solutions for patients. In a more recent development, the company was an early adopter of Generative AI to power iQNow, an in-house platform created to help scientists at Boehringer search through large collections of research papers. In just the first nine months after its introduction in early 2023, the company combined iQNow with the interactive features of large language models. This enhancement propelled the number of unique users at the company from 1,000 to 25,000.
How did Boehringer manage to become a forerunner in using Generative AI?Boehringer Ingelheim did its homework and was well prepared. Several years ago, when multiple departments at the company began reporting they were overwhelmed with information, Boehringer Ingelheim sought out a solution to improve access to scattered documents and multiple databases compiled according to varying standards. The company saw forthcoming advances in AI as a path forward. IQNow was designed to help retrieve specific information using a sophisticated algorithm, and it was combined with AI-driven large language models when they became available.
Are there risks associated with connecting the company's proprietary data to an AI application that lives in the cloud?To ensure that proprietary data is kept fully secure, iQNow was connected to Microsoft's Azure Open AI Service as a safe way to bring these complementary technologies together. Full separation is maintained between public and private data sources.
How is iQNow changing Boehringer's research?Boehringer Ingelheim's researchers use iQNow to comb through roughly 700 million document databases to quickly identify experts on specific areas of study. Researchers ask iQNow questions as if it is a colleague sitting next to them who can give a reliable answer. By following the logic of human conversation, it helps them break down any research questions they encounter. Researchers can proceed to ask deeper questions or request supporting documents if they want more information.
Has Generative AI measurably helped with research at Boehringer Ingelheim?Yes, Generative AI has measurably helped Boehringer Ingelheim's research. After its initial launch, a select group of staff members used iQNow to supplement their research. But with the release of ChatGPT, generative AI became a more practical, everyday tool, and the company combined iQNow with the interactive features of a large language model. Users reported that, in the first 70 days after this system came online, it saved nearly 150,000 hours of labor. In the first nine months, it helped save approximately 600,000 working hours, equivalent to roughly €60 million in research spending.
Besides just saving time, how has this practical application of AI helped the company's research?Rather than just steering Boehringer Ingelheim researchers to specific documents or experts, AI now gives them the ability to engage in a moderated dialogue drawing on all the knowledge accessible from across a vast database of research documents. This contributes directly to accelerating Boehringer Ingelheim's process of drug discovery and development.
What else has been going on with AI at the company?The company applies AI in a variety of ways. What started with iQNow, Generative AI for everyone, is meanwhile realized with Microsoft standard applications such as Copilot. Other examples demonstrate the effectiveness of Generative AI for specific business use cases: AI is now being leveraged by Boehringer Ingelheim to summarize insights captured by medical-scientific liaison teams. The company's first product in this field, KNERD, is now active and being used by researchers. ADAM, the company's own advanced design assistant for molecules, empowers medical chemists in their hunt for new drug candidates. |
These cases demonstrate the value as well as the heterogeneity of AI usage across Boehringer Ingelheim. Meanwhile, Boehringer's IT has set-up a robust solution framework, which offers standard AI solutions for all users as well as capabilities and platforms to develop and run custom solutions for very specific use cases in the businesses and organizational functions.
How is AI expected to affect the pharmaceutical industry in the future?
AI is expected to transform the pharmaceutical industry in several ways. It will accelerate drug discovery, speeding up the process of identifying compounds and targets for possible treatments. This will lead to new drugs that might never have been developed without this remarkable tool, for example with molecule structure prediction or de-novo protein design.
It will soon be capable of supporting human work in low-risk clinical trial settings, leading to quicker regulatory approvals and faster availability of new treatments for patients who need them.
Furthermore, AI will accelerate the development of personalized medicines tailored to individual patients – something that will ultimately improve treatment outcomes.
To explore the potential of quantum computers for pharmaceutical research and development, Boehringer Ingelheim launched a collaboration with Google Quantum AI in 2021. What is the current state of advancement in this field?
Three questions for Clemens Utschig-Utschig CTO & Chief Architect IT, Boehringer Ingelheim
1. Where do we currently stand in terms of using quantum computing for pharmaceutical research?
Quantum computing may be on the brink of enabling highly accelerated computing power. And chemistry may prove to be the first major use case. Computational drug discovery relies on making accurate predictions of how candidate drugs will interact with their targets in living cells. This requires the simulation of thousands of atoms at specific temperatures, and it can generate insights into these systems that may otherwise be inaccessible through traditional experimentation.
Quantum computers promise to perform highly efficient chemical calculations that can simulate the quantum nature of the system. These tasks, involving complex molecules, are beyond the capabilities of conventional computers. Today's quantum computers are still at an early stage of development and can only be used for very small systems. To simulate pharma-relevant molecules we’ll need larger and more reliable quantum computers, as well as quantum algorithms tailored to our needs. A central goal is to make computer modeling equivalent or more convenient than current lab experiments. Quantum hardware is advancing very quickly, and novel quantum algorithms are being developed every day. So practical applications are getting closer.
2. What has Boehringer Ingelheim learned so far? And when will using quantum computers become a standard tool that can start to benefit patients?
To explore quantum computing's potential, Boehringer Ingelheim launched a collaboration with Google Quantum AI at the beginning of 2021.
Within this collaboration, we have explored several paths forward to practical applications. For example, one of our use cases was to identify quantum algorithms to study the P450 enzyme. P450 plays an important role in the human metabolism and has never been analyzed this way before. The outcome of the analysis has shown that quantum computers can offer a clear advantage over the best classical methods at very high level of accuracy.
However, even with the best available quantum algorithms, these calculations would require three days of runtime, which is way beyond what is practical in an industrial setting. We are currently working on developing new algorithms that could reduce computer runtimes from hours or days to a few minutes.
Another example of our current research, together with the University of Toronto, involves developing quantum algorithms to study molecular dynamics, a field that seeks to predict how molecules move over time.
Our key goal is to predict how well drug candidate molecules will bind to their target. Therefore, we have developed a novel quantum algorithm for molecular dynamics and have presented those results at various international conferences.
Nonetheless, while we are making steady progress in terms of software, hardware and use cases, we are still at the stage of applied basic research.
Further development will be needed to exploit the economic potential of quantum computing. We are continuing to push this forward and expect to be able to point to examples of industry-relevant applications by the end of this decade.
3. What are the next steps that need to be taken?
It's still too early to predict when the pharmaceutical industry will be able to harness the full potential of quantum computers.
We need to see further improvements in hardware and the development of novel algorithms. We also need to come up with new methods that allow us to make compromises between accuracy and the amount of time needed for calculations.
The main focus for the time being will be to keep reducing the runtimes of quantum algorithms – to the point where these calculations will be more attractive than either experiments or low-accuracy calculations from conventional computers – all while exploring new use cases.
In other words, there are many challenges that we, together with our partners, can actively contribute our expertise to solving. I am certain that the next years will lead up to the advancements we need.
The perspective of our quantum lab and partners has been published here in Nature Physics.