Catch up on the latest AI articles

[DeepCRE] Cutting-Edge Computational Models Revolutionize Drug Research And Development

[DeepCRE] Cutting-Edge Computational Models Revolutionize Drug Research And Development

Machine Learning

3 main points
✔️ Introducing DeepCRE, a new computational model that addresses the problem of inadequate drug-to-drug response evaluation (CRE) in late-stage drug development
✔️ DeepCRE achieves an average 17.7% performance improvement in patient-level CRE and a fivefold increase in indication-level CRE, outperforming existing models outperforms

✔️ Proposed future use of LLM to analyze genomic expression profiles of patients and to expand the drug R&D process

DeepCRE: Revolutionizing Drug R&D with Cutting-Edge Computational Models
written by Yushuai WuTing ZhangHao ZhouHainan WuHanwen SunchuLei HuXiaofang ChenSuyuan ZhaoGaochao LiuChao SunJiahuan ZhangYizhen LuoPeng LiuZaiqing NieYushuai Wu
(Submitted on 6 Mar 2024)
Comments: Published on arxiv.

Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Quantitative Methods (q-bio.QM)

code:  

 

The images used in this article are from the paper, the introductory slides, or were created based on them.

Summary

The field of drug development and therapeutic applications faces significant challenges. More alternative treatments are needed in the therapeutic area, yet many promising drugs fail in clinical trials. One reason for this is inadequate drug-to-drug response evaluation (CRE) in the late stages of drug development. A new computational model to address this was introduced, DeepCRE, which outperforms the best existing models by achieving an average 17.7% performance improvement in patient-level CRE and a 5-fold increase in indication-level CRE. In addition, DeepCRE identified drug candidates with significantly higher efficacy in colorectal cancer (CRC) organoids. This highlights DeepCRE's ability to find drug candidates with superior therapeutic efficacy and its potential to revolutionize the field of therapeutic development.

DeepCRE

The model integrates patient gene expression profiles and cell line data with improved pre-training strategies, allowing for highly accurate prediction of drug response. In particular, adjustments focused on patient type played an important role in building the model. This improved the accuracy of drug response predictions and led to improved subsequent performance.

Performance

The DeepCRE performance evaluation compared models in different tumor types. In particular, the DSN-adv model performed better than the other models. On average, this model achieved performance gains of 27.49%, 21.38%, and 17.08%. DeepCRE also demonstrated its usefulness in evaluating the efficacy of drug candidates and validating their reuse.

The Potential of DeepCRE in Assessing the Clinical Drug Value of Preclinical Drug Candidates

DeepCRE is useful for assessing the clinical drug value of preclinical drug candidates. This assessment evaluated 233 small molecules across 13 tumor types and was validated using the DrugBank, ClinicalTrials, and Repurposition Hub databases.

Evaluation results showed that DeepCRE was able to identify efficient drug candidates; the DEI table showed predicted efficient drug candidates, and these predictions showed excellent agreement with clinical trial records. In addition, many of the drug candidates identified by DeepCRE were validated in real clinical trials, finding more qualified drug candidates than previous SOTA models.

Results show that DeepCRE is practical in assessing pharmacological value at the preclinical stage and is consistent with the expertise of real-world professionals.

Validation of DeepCRE for efficient drug candidate identification in CRC21 patients

DeepCRE's capabilities were validated in the identification of efficient drug candidates in a clinical setting focused on CRC21 patients. This validation used tumor samples from patients who had experienced recurrence despite XEOLX treatment.

Drug candidates identified by DeepCRE were categorized into four sets based on their MoA (mechanism of action) and rating differences from traditional methods. This ensured that each drug set maintained MoA diversity and the drug set with the greatest rating disparity.

Results show that DeepCRE outperforms traditional methods. First, the drug candidates identified by DeepCRE demonstrated significantly higher efficacy than a comparative set of approved drugs. Furthermore, drug candidates with specific MoAs show potential to outperform conventional chemotherapy. These results demonstrate that DeepCRE is a powerful tool for identifying effective treatments tailored to individual patient characteristics.

Drug repurposing validation of DeepCRE in 8 CRC organoids

Seven additional CRC organoids were tested in the validation of DeepCRE for drug repurposing, for a total of eight CRC organoids. The results showed that for five CRC organoids, the drug candidates identified by DeepCRE showed a significant increase in efficacy over Comparison Set C.

Efficacy of drug candidates was consistent among different patients based on MoA. In particular, inhibitors of PI3K/mTOR signaling and chromatin histone acetylation showed significant increases compared to drugs with other MoAs.

In addition, the results of the drug trials provided several interesting insights. For example, treatment with certain drugs showed upward regulation of cell survival and stress response pathways. In addition, the role of transport proteins may influence drug resistance.

Molecular docking analysis showed binding affinity to transport proteins targeted by specific drugs. In addition, a negative correlation between ABCG2 expression and drug resistance in CRC was revealed.

These results indicate that DeepCRE is a powerful tool for predicting drug effects among different patients and identifying drug candidates that may be repurposed.

Conclusion

This study demonstrates the proposed DeepCRE model and its revolutionary potential in the drug R&D process: DeepCRE outperforms the traditional SOTA model through advances in pre-training strategies, achieving an average performance improvement of 17.7% at the patient level and a 5-fold increase at the indication level This is a tremendous increase. In addition, DeepCRE identified drug candidates with significant efficacy among the CRC organoids tested.

It also reveals pharmacological and pharmacodynamic insights about specific drug candidates that may inform future research and development processes. Importantly, DeepCRE emphasizes the ability to discover a collection of drug candidates with enhanced efficacy, rather than simply identifying one or two potential drugs.

Suggested future directions to consider include the analysis of patients' genomic expression profiles (GEPs) using large-scale language models (LLMs), the expansion of the drug research and development process to include a generative paradigm, and the establishment of synergistic approaches that integrate computation and experimentation. These approaches are expected to facilitate the development of effective and creative therapies that will benefit patients.

 
  • メルマガ登録(ver
  • ライター
  • エンジニア_大募集!!

If you have any suggestions for improvement of the content of the article,
please contact the AI-SCHOLAR editorial team through the contact form.

Contact Us