Embracing artificial intelligence design for better radiopharmaceuticals

Abstract

Cancer has emerged as a significant threat to human life, and its incidence and mortality are increasing rapidly. As clinicians increasingly seek to noninvasively investigate tumor phenotypes and evaluate functional and molecular responses to therapy, the combination of diagnostic imaging with targeted therapy is becoming more widely implemented [1]. Targeted radionuclide therapy involves the use of small molecules, peptides, and/or antibodies as carriers for therapeutic radionuclides, with these being referred to as radiopharmaceuticals. Radiopharmaceuticals, also known as molecular probes in nuclear medicine, play a vital role in clinical diagnosis and therapy. Currently, there are numerous radiopharmaceuticals approved or under research worldwide for a wide range of indications. At the end of March 2022, there were 60 radiopharmaceuticals approved for marketing by the Food and Drug Administration (FDA) [2] (Supplemental Table) [2]. As of October 2023, 42 radiopharmaceuticals have been approved for marketing by the National Medical Products Administration (NMPA) [3]. However, there remains an urgent need to identify new targets and new drug molecules to advance the process of radiopharmaceutical research and development.

In 2022, the nanobody ⁶⁸Ga-Nb1159 targeting the receptor-binding domain (RBD) of Severe Acute Respiratory Syndrome Coronavirus type 2 (SARS-CoV-2) [4], was successfully prepared. The probe has the potential not only to monitor the distribution of SARS-CoV-2 in real time, but also to assess the infection status of patients. However, its targeting specificity is limited by the structural characteristics of the small molecule. Therefore, designing radiopharmaceuticals with high specificity and affinity has become an important direction in the development of radiotherapy drugs.

The development of artificial intelligence (AI) has brought new technological tools for drug discovery and development. New possibilities for the design, synthesis, and bioanalysis of new and existing small molecules have been opened up through machine learning (ML), deep learning (DL), and so on [5]. For example, AI can be applied to the design of ligands with high affinity for research targets. Such ligands for radiopharmaceuticals can be nanoparticles; however, unlike the nanoparticles traditionally used to deliver chemotherapeutic drugs, AI-driven nanoparticles that can respond to or be guided by biological cues are emerging as a promising drug delivery platform for the precise treatment of cancer. A study of AI-guided polymer nanoparticles showed that the fluorescence intensity and wavelength generated by the interaction between negatively charged cyclic peptide nanoparticles and amyloid-beta aggregates in cerebrospinal fluid and serum varied with disease state in comparison withed to healthy individuals [6, 7]. The proposed computer-aided design of smart nanoparticles further enhances the potential of these ingenious nanoscale technologies to provide personalized treatment options for patients.

Ligands for radiopharmaceuticals can also be active substances such as peptides and proteins. Among them, the structure of proteins is the most complex because the arrangements and spatial structures of the amino acid sequences that make up proteins are very complex, and the number of possible sequence combinations is enormous. Therefore, the technology of using amino acid sequences to predict protein structures and synthesize proteins is poised to significantly accelerate the development in life sciences. For example, an independent analysis conducted by Spanish researchers [8] showed that the AlphaFold algorithm reduced the number of human proteins without structural data from 5027 to 29.

Designing peptides from scratch on the basis of protein sequences is a challenging process. Thomson et al. synthesized target peptides for 22 targets and predicted the oligomer-state using CCBuilder software, which successfully predicted the states of 8 out of 13 peptides [9]. However, while this method contributes to stability, it does not ensure the oligomer-state specificity of the peptides. Furthermore, Gevorg et al. presented a computational framework for designing protein interaction specificity and demonstrated its application by identifying specific peptide partners for the human basic zone leucine zip (bZIP) transcription factor [10]. bZIP proteins presents challenges due to their strong sequence and structural similarities. The computational method offers a trade-off between stability and specificity, potentially serving as a versatile tool for peptide design.

In this article, we elucidate the fundamental concepts related to radiopharmaceuticals for diagnostic or therapeutic applications, shedding light on the specific challenges encountered in the design, ex vivo and in vivo evaluation, and translation of novel radiopharmaceuticals. The emergence of AI tool, such as AlphaFold, has led to a number of breakthroughs in the life sciences. Prediction of the function and design of proteins has been one of the first areas to benefit. Perhaps this technology can be used in the design of radiopharmaceutical carrier molecules, and in the near future, we will see radiopharmaceuticals based on this new design ideal for the benefit of patients.

Radiopharmaceuticals are radiolabeled formulations or precursors used in the practice of nuclear medicine for diagnostic, therapeutic, and disease surveillance purposes, as well as for research tools in the pharmaceutical industry [11]. They are usually classified into diagnostic and therapeutic radiopharmaceuticals. With advances in radioimmunotherapy and radioligand drug therapy, China's radiopharmaceutical market is expanding rapidly. However, the supply of radionuclide and nuclear medicine imaging equipment in China largely relies on imports, and the amount of available equipment is relatively small, seriously restricting the development of China's radiopharmaceutical industry. In order to reduce the risks when precursors are used for patients and reduce the consumption of medical resources, new technologies are urgently needed to solve the problem of this “bottle neck” in both China and worldwide, to stimulate research and development.

Preclinical evaluation is an integral part of the radiopharmaceutical development. Over the years, advances in biology and chemistry-related disciplines have led to the use of various molecules to develop a new generation of radiopharmaceuticals whose purpose is to deliver radioisotopes to specific targets at the cellular or molecular level. This necessitates a thorough evaluation of radiolabeled molecules during the preclinical stage to assess their safety and suitability for the intended clinical application. These studies involve measurements of stability, affinity, and target-specificity, where affinity and targeting specificity are primarily influenced by ligands (such as antibodies, peptides, and nanoparticles, etc., that bind specifically to the target) linked to radionuclides. Therefore, the structural design and modification of ligands are key elements to solving the problem of radiopharmaceutical deficiency.

AI continues to make breakthroughs in predicting the three-dimensional structure of biological molecules, and guiding the synthesis of novel radiopharmaceuticals (Figure 1). De novo protein design first appeared around the 1980s [12]. By the end of 2020, AlphaFold, a DeepMind-developed deep-learning algorithm, could potentially solve the problem of “protein folding” and the prediction of different protein structures, which had eluded academia for decades. With the emergence of AlphaFold2 in CASP14 (14th Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction), the problem of protein folding can be said to have been basically solved, and deep learning has completely changed the field of protein structure prediction.

AlphaFold2 improves the speed of protein structure resolution, and the system can generate accurate protein structures in minutes to hours [13]. David Baker's team at the University of Washington published RoseTTAFold, a protein prediction design method of comparable accuracy, in Science, which can analyze the 3D structure of a given sequence in less than 30 minutes [14].

Using the protein design software Rosetta, David Baker's team at the University of Washington's Institute for Protein Design [15] has created a new protein, Neo-2/15, that can imitates the action of IL-2, and has shown superior therapeutic activity to IL-2 in mouse models of melanoma and colon cancer, with reduced toxicity and undetectable immunogenicity. Even after 1 h of incubation at 95°C, Neo-2/15 effectively promotes T cell survival, whereas native IL-2 can no longer remain active under these conditions. This achievement opens up a broad viewpoint for cancer therapy, autoimmune diseases and other diseases based on engineered proteins. Furthermore, the temperature tolerance of the designed structure may stimulate further modifications of itself.

ChatGPT and other Large Language Models also contributes to drug target discovery. As described in a news article in Nature Biotechnology, ChatGPT can provide important information on drug target discovery for specific diseases. Zhao et al. describe how ChatGPT can help in the development of innovative drugs, predicting different characteristics of drugs, such as molecular pharmacodynamics, pharmacokinetics, and toxicity [16]. Recently, Ali Madani et al. described the ProGen language model [17], which can generate protein sequences with predictable functions in large protein families. This model was trained on 280 million protein sequences and enhanced with control tags that specify protein properties. Their study shows that recent advances in deep learning-based language modeling can be employed to generate, from scratch, artificial protein sequences that function like natural proteins.

There are still limitations in studying this aspect of targeted molecular imaging and radio-ligand therapy: 1. Insufficient binding affinity; 2. Target screening and molecular optimization are difficult and therefore bio-functional chelator and nuclides should be brought into AI computing together; and 3. Temperature sensitivity is a challenge for proteins.

AI plays a pivotal role in radiopharmaceutical development by excelling in two crucial aspects. First, it can accurately predict the structure of the designed protein. Second, on this basis, and according to the orientation of the amino acid side chain of the binding site, a protein that can be accurately bound to it is designed. In addition, the evaluation of the obtained protein is also very important. There are broad applications for and several great advantages to be had from introducing AI technology into radiopharmaceutical design: 1. Better binding affinity, even to pM (pmol/L) level; 2. Ligands, chelators, and radionuclides can be brought into AI computing together; 3. AI technology can show super structural stability, with stability at temperatures as even high as 95°C. More importantly, radiopharmaceuticals can be prepared by prokaryotic rapid expression, which significantly reduces production costs.

Recent advances in AI, including the development of more sophisticated ML techniques, have had a dramatic impact on the drug discovery process. In this article, we highlight recent typical applications of AI in drug discovery, illustrating the increasingly important role AI plays in drug design. Furthermore, experimental validation of the effectiveness of AI in drug development is a key factor in understanding how AI contributes to medicinal chemistry and how it can be further developed and improved [18]. The application of AI in this phase needs to be supported by scientists and therefore it is extremely important to draw their attention to its potential.

AI technologies, including deep learning and machine learning, have been widely used in the medical field. The advantage of these technologies is that they can extract useful information from large amounts of data to predict and optimize experimental results. In the development of radiopharmaceuticals, AI technologies can help us better understand the nature of radiopharmaceuticals and predict their possible effects, thus accelerating the development of new drugs. In addition, the AI technology can help us understand the mechanism of action of radiopharmaceuticals. By analyzing the data on drug-target interactions, we can better understand the principle of action of a drug and thus optimize its design and use.

Overall, the application of AI technology has great potential in the radiopharmaceutical research and development. It can solve the current problems of the limited number and insufficient innovativeness of radiopharmaceuticals and improve the efficiency of the research and development of new drugs. To promote the advancement of nuclear medicine, it is imperative to intensify research and development efforts by leveraging AI technology within radiopharmaceutical research and development.

Jinping Tao performed the manuscript editing and preclinical investigations; Xiangxing Kong contributed through a grammatical revision of the manuscript; Zhi Yang and Hua Zhu contributed through supervision and manuscript editing. All authors read and approved the final manuscript.

The authors declare no conflict of interest, if authors are from the editorial board of RADIOLOGY, they will be excluded from the peer-review process and all editorial decisions related to the publication of this article.

This article does not include any original work with human or animal subjects performed by any of the authors.