Introduction

Explor Target Antitumor Ther

ETAT

Exploration of Targeted Anti-tumor Therapy

2692-3114

Open Exploration Publishing

10.37349/etat.2023.00197

1002197

Review

Current implications and challenges of artificial intelligence technologies in therapeutic intervention of colorectal cancer

Das

Kriti

Conceptualization Investigation Writing—original draft Visualization ¹ ^† Paltani

Maanvi

Conceptualization Visualization ¹ Tripathi

Pankaj Kumar

Conceptualization Investigation Writing—original draft Visualization ² ^† Kumar

Rajnish

Writing—review & editing Supervision ³ Verma

Saniya

Writing—review & editing Supervision ³ Kumar

Subodh

Writing—review & editing Supervision ³ ^* Jain

Chakresh Kumar

Conceptualization Writing—review & editing Supervision ² ^* Nardone

Valerio

Academic Editor University of Campania, Italy

¹Department of Artificial Intelligence and Precision Medicine, School of Allied Health Sciences and Management, Delhi Pharmaceutical Sciences and Research University, New Delhi 110017, India ²Department of Biotechnology, Jaypee Institute of Information Technology, Noida 201309, Uttar Pradesh, India ³Department of Medical Laboratory Technology, School of Allied Health Sciences, Delhi Pharmaceutical Sciences and Research University, Delhi 110017, India

†

These authors contributed equally to this work.

^*Correspondence: Subodh Kumar, Department of Medical Laboratory Technology, School of Allied Health Sciences, Delhi Pharmaceutical Sciences and Research University, Delhi 110017, India. skumarims07@gmail.com Chakresh Kumar Jain, Department of Biotechnology, Jaypee Institute of Information Technology, Noida 201309, Uttar Pradesh, India. ckj522@yahoo.com

2023

27 12 2023

4 6 1286 1300 12 06 2023 28 08 2023

This is an Open Access article licensed under a Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, sharing, adaptation, distribution and reproduction in any medium or format, for any purpose, even commercially, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Irrespective of men and women, colorectal cancer (CRC), is the third most common cancer in the population with more than 1.85 million cases annually. Fewer than 20% of patients only survive beyond five years from diagnosis. CRC is a highly preventable disease if diagnosed at the early stage of malignancy. Several screening methods like endoscopy (like colonoscopy; gold standard), imaging examination [computed tomographic colonography (CTC)], guaiac-based fecal occult blood (gFOBT), immunochemical test from faeces, and stool DNA test are available with different levels of sensitivity and specificity. The available screening methods are associated with certain drawbacks like invasiveness, cost, or sensitivity. In recent years, computer-aided systems-based screening, diagnosis, and treatment have been very promising in the early-stage detection and diagnosis of CRC cases. Artificial intelligence (AI) is an enormously in-demand, cost-effective technology, that uses various tools machine learning (ML), and deep learning (DL) to screen, diagnose, and stage, and has great potential to treat CRC. Moreover, different ML algorithms and neural networks [artificial neural network (ANN), k-nearest neighbors (KNN), and support vector machines (SVMs)] have been deployed to predict precise and personalized treatment options. This review examines and summarizes different ML and DL models used for therapeutic intervention in CRC cancer along with the gap and challenges for AI.

Artificial intelligence machine learning deep learning colorectal cancer drug discovery

Introduction

Cancer is the second leading cause of death across the globe [1]. In terms of mortality and morbidity, Global Cancer Statistics 2020 shows that, out of 36, colorectal cancer (CRC) is the third most common cancer in the population. Worldwide, it affects equally to both men and women equally. Every year more than 1.85 million cases of CRC have been reported and 20% of them have metastatic disease at presentation. The estimated number of deaths by 2023 for CRC is 52,550 [2]. CRC is the third most common type of cancer in both sexes and demands an early diagnosis and treatment to save the lives of many [3]. It begins with the formation of tiny clusters called polyps. Some of these polyps turn malignant resulting in CRC over a period of 10–15 years. Males are more likely to be associated with CRC. Family history of CRC in correspondence to age and the extent of its effect on the relatives play a role in 10–20% of all CRC patients. Individuals with older age and lasting bowel inflammation are at higher risk [4]. Researchers of various domains are discovering different approaches to tackle this disease.

Artificial intelligence (AI) can be used to improve CRC treatment and treatment methods. With the huge amount of data generated by medical imaging, computed tomography, histopathology evaluation, etc. comes the use of AI [5].

In addition, machine learning (ML) algorithms can be used to create predictive models to help clinical decision-making without any prior explicit programming [5]. Many modalities and sub-specialties of AI show promise for the application of predictive studies, distribution, and prevalence of CRC and thus enable personalized approaches in drug discovery uplifting precision medicine and subsequently clinical practices [6, 7]. Drug development is inevitably a delicate and challenging procedure that puts a strain on productivity and research and development (R&D) costs.

The diagnosis and categorization of diseases and their subtypes among patients are made possible by a variety of deep learning (DL) and statistical techniques that depend on data interpretation.

To detect disease targets quickly and accurately, ML, feature-finding, and clustering techniques are useful. The application of statistical analysis on big data, experimental data, and data mining methods, together with neural networks, improves capacity for de novo drug designing (DD). The use of existing drugs for new therapeutic applications is drug repurposing also called repositioning. The use of metformin, a type 2 diabetes medication showed reduced chances of developing CRC in 47,000 participants [8]. The area of precision medicine is advanced by drug repurposing and combination therapy based on numerous genomic markers and increased patient information (Figure 1) [6]. Based on the literature survey, this review provides a description of CRC and the performance of various AI-based models in chemotherapy and neoadjuvant chemoradiotherapy (nCRT) for its therapeutic intervention. It also shows the application of existing AI-backed computational tools in the domain of drug discovery and development processes to enhance treatment options for colon cancer and other disease conditions. Finally, it spreads light on the challenges that lie ahead of AI for drug repurposing, de novo DD, and therapeutic options for various disease conditions including CRC.

Figure 1

How and where AE in cancer research is being used

Functionalities of AI

A subfield of computer science called AI encompasses various fields, including mathematics, logic, philosophy, psychology, cognitive science, and biology. AI refers to intelligent technology that has been artificially created to mimic humans. This AI is incorporated into a computer system known as an AI system, which eventually serves as a thinking machine. The three characteristics of the AI system are intelligence, intentionality, and adaptability. A variety of strategies can be used to create an AI system that effectively performs human tasks. AI supports the system’s decision-making process and aids in outcome forecasting. To advance current technology or develop new ones, AI combines different ML algorithms and neural networks (Figure 2).

Figure 2

Applications of AI in different disciplines, utilizing DL and ML

Basics of ML

A subtype of AI called “ML” enables computers to learn from their surroundings automatically and without human involvement, which suggests that they are developing their decision-making abilities. ML employs a number of algorithms and strategies to classify and enhance the data to make better predictions. In the medical sciences, ML techniques are now applied for the detection and categorization of distinct tumor forms. First, ML algorithms look for patterns, and then they take actions based on those patterns [9, 10]. ML can be primarily divided into three types: supervised learning, unsupervised learning, and reinforcement learning (Figure 3).

Figure 3

Categorization of ML algorithms with its subtypes and their applications

In supervised learning, both the input and output data are provided by the trainer. It is a kind of ML in which the data has been labeled so that the machine may discover and build patterns between the input and output data. By identifying the pattern, it can learn how to classify or categorize the data [11]. Supervised learning is of two types viz (A) classification and (B) regression.

Classification is a type of supervised learning that is used to predict/classify discrete values such as male or female, yes or no, malignant or benign, etc. Some classification algorithms under supervised learning are decision trees, random forest (RF), logistic regression (LR), support vector machines (SVMs), etc. [12].

Similarly, regression is another type of supervised learning used to predict continuous values. Some regression algorithms under supervised learning, i.e., regression trees, linear regression, non-linear regression, polynomial regression, etc. [12, 13].

Unsupervised learning is a type of ML in which algorithms may uncover previously undiscovered patterns in unlabelled datasets and provide the desired output without any external help. Unlabelled datasets are analyzed and clustered using ML techniques [11]. Unsupervised learning is of two types viz (A) clustering and (B) association. Clustering is a technique for organizing data points into various clusters made up of related data points. Finding correlations between variables in a large database is done using the unsupervised learning technique of association.

Some algorithms are under unsupervised learning, i.e., k-means clustering, apriori algorithm, hierarchal clustering, independent component analysis, k-nearest neighbors (KNN), and principle component analysis (PCA).

Reinforcement learning is an ML strategy that relies on feedback, in which an agent learns automatically utilizing feedback rather than labeled data. The agent is necessitated to learn exclusively from its own experience because there isn’t any labeled data.

SVM

Researchers have used ML techniques for datasets for the diagnosis and treatment of cancer. Although there are many methods proposed for classification, SVM is the most popular due to its strong mathematical foundation based on structural risk minimization, statistical learning theory, and its accurate performance. SVM is a pattern recognition tool [14]. SVM is being used in many ways in the field of drug discovery (Figure 4) [15].

Figure 4

Applications of SVM in drug discovery

For linear classification, the decision function can be formulated as equation 1 (Eqn. 1). For non-linear classification, the decision function can be formulated as Eqn. 2.

Eqn. 1:

fx = sign∑i = 1pyiαix.xi + b

Eqn. 2:

fx = sign∑i = 1pyiαikx.xi + b

Where, k = kernel function; x_i = n-dimensional vector; y_i = its label; and a_i = Lagrange multipliers.

Naive Bayes classifier model

Naive Bayes classifier (NBC) are simple “probabilistic classifiers” based on Bayes theorem with naive (strong) independence assumption between the features (Eqn. 3). Naive Bayes has been used to diagnose CRC by identifying the origin of tumor cells using RNA sequence data [16].

Eqn. 3:

Pcx = PcPxcPx

Where, x = attributes; c = class; P(c|x) = probability of “c” being true, given that “x” is true; P(x|c) = probability of “x” being true, given that “c” is true; P(c) = probability of “c” being true; and P(x) = probability of “x” being true.

Using Bayesian probability, the above Eqn can be written as Eqn. 4.

Eqn. 4:

Posterior = Prior × LikelihoodEvidence

LR is the practical application of AI for disease prognosis and management. LR models predict the probability of values ranging from 1 and 0. It is mostly applied to categorical data [17]. For example, if the cancer is malignant (1) or not (0). LR can be represented by Eqn. 5:

Eqn. 5:

log⁡y1 - y = b0 + b1x1 + b2x2 + b3x3 + … + bnxn

Where, y = predicted output; b₀ = intercept; x = input value; and b₁ = co-efficient of the input value 𝑥 (single value).

DL is an ML technique that trains a computer to filter inputs through layers as it gains the ability to predict and categorize data. It basically consists of a neural network with three or more layers. These neural networks attempt to mimic how the human brain functions [12, 18]. This has been classified broadly into convolutional neural networks (CNNs), artificial neural networks (ANNs), and recurrent neural networks (RNNs).

CNNs are a specific type of neural network that is mostly used for object recognition, image clustering, and image classification [19], while the ANNs mimic the biological neural networks of the human brain and are typically comprised of three layers, i.e., (A) the input layer which accepts input from the programmer in a variety of formats; (B) hidden layer: these layers are situated in-between the input and output layers and it performs all calculations to reveal hidden characteristics and patterns; and (C) output layer: this layer is used to convey the output after the input has undergone several transformations utilizing hidden layers.

RNNs are a specific kind of ANN that is mostly used in speech recognition and natural language processing (NLP). Because their mathematical processes are performed sequentially, RNNs get their name [19, 20].

Steps of drug discovery and AI

The long and difficult process of finding new drugs can be roughly broken down into the following stages: (A) target identification; (B) target validation; (C) lead identification; (D) lead optimization; (E) product characterization; (F) formulation and development; (G) pre-clinical research; (H) investigational new drug; (I) clinical trials; (J) new drug application; and (K) approval [21]. Regarding a certain ailment, it is required to first determine the target. In the following phase, hit identification, molecules in molecular libraries are identified using techniques including combinatorial chemistry, high-throughput screening, and virtual screening. In a clinical study, the medication candidate is finally given to patients after passing all preclinical tests satisfactorily.

The medication must proceed sequentially through each of the three stages of this process. Phase I entails doing drug efficacy tests on a handful of individuals with the specified ailment; phase II entails running drug safety tests on a smaller number of human subjects; and phase III entails performing effectiveness tests on a wider range of patients. If the drug candidate’s safety, as well as effectiveness, are shown during the clinical phases, agencies like the Food and Drug Administration (FDA) review the substance for authorization and marketing. A traditional drug discovery pipeline is thought to cost an average of 2.6 billion dollars, and it may take up to 12 years to accomplish [22, 23].

The main concerns for all pharmaceutical companies are how to save expenses and advance initiatives. To increase productivity and cut costs, AI-based computational tools are being used at various phases of the drug discovery process (Table 1). These include cell classification and real-time image-based cell sorting, as well as computer-aided organic synthesis, design of new molecules, assay development, and prediction of the three-dimensional (3D) structures of target proteins, among many other uses (Figure 5). In general, AI can automate and optimize these time-consuming processes to dramatically speed up R&D medication development [24, 25]. Also, AI is used to coordinate, operate, and recruit participants for clinical trials, frequently associated with improved patient monitoring during clinical trials or with medical equipment that can access specific patient data and guide medical decisions [26, 27].

Table 1

Computational tools for drug discovery: AI-based

Tools	Purpose	References
DeepChem	Drug discovery task prediction	[28]
DTI-CNN	DL based drug-target interaction prediction	[29]
ORGANIC	Molecular generation tool with desired properties	[30]
Chemputer	Chemical synthesis reporting procedure	[31]
DeltaVina	Rescoring protein-ligand binding affinity: scoring	[32]
DeepCPI	Drug–protein interaction prediction	[33]
PotentialNet	A CNN graph-based ligand-binding affinity prediction	[34]
DeepNeuralNet-QSAR	Prediction of molecular activity	[35]
Hit Dexter	Prediction of molecules responding to biochemical assays	[36]
DeepTox	For toxicity prediction	[37]
PPB2	Polypharmacology prediction	[38]
SCScore	For evaluation of the synthesis complexity of a molecule	[39]
NNScore	Protein-ligand interaction scoring study	[40]
SIEVE-Score	Structure-based virtual screening	[41]
REINVENT	Molecular de novo design based on RNN and RL	[42]

RL: reinforcement learning; DTI-CNN: drug-target interaction-CNN; QSAR: quantitative structure-activity relationship; PPB2: polypharmacology browser 2; SCScore: synthetic complexity score; SIEVE-Score: similarity of interaction energy vector-score; DeepTox: DL for toxicity; NNScore: neutral-network receptor-ligand scoring function

Figure 5

AI in drug screening, DD, drug repurposing, and chemical synthesis

A quick overview of the recent instances of drug development using AI techniques has been discussed as shown in Figure 5. The developing field of AI has garnered limited attention despite its significant expansion. The computational creation of novel structures with desired attributes, known as de novo design, is a focal point, particularly starting with fresh chemical matter. Likewise, the related domains of forward prediction and retrosynthesis prediction, seeking to establish how chemical matter designated for experimental research can be synthesized, have also piqued substantial interest. Determining whether a ligand binds to a specific protein target once it has been placed is the logical next step, and target prediction in silico and docking (and related techniques) have been active research fields for decades [43]. In terms of predicting ligand-protein interactions, methods like DL have a somewhat good effect on improving numerical measurements of performance (often marginally). This hasn’t always been the case, though, as evidenced by a recent large-scale study that found no benefit to DL in terms of performance. Also, special attention must be paid to the model performance measurements employed in this context and if they reflect a pertinent metric capable of detecting both significant and practically applicable changes in model quality [44].

AI in drug discovery

The process of creating efficient new pharmaceuticals is the most complicated part of the medication development process. The techniques that incorporate AI have evolved into flexible toolkits that can be used widely in several stages of drug development, including the identification and validation of drug targets, the design of new drugs, drug repurposing, improving R&D efficiency, the analysis of biomedicine data, and the improvement of the decision-making process to enroll patients in clinical trials [21]. While addressing the inefficiencies and uncertainties brought on by the conventional techniques of drug development, these potential applications for AI offer the ability to reduce bias and human meddling in the process [45, 46].

Further applications of AI in drug development include pharmacological qualities, protein features and efficacy, drug combination and DTI, drug repurposing, drug synergism/antagonism prediction, and prediction of practical synthetic methods for drug-like compounds [47]. Finding new pathways and targets using omics research is made feasible by the development of novel biomarkers and therapeutic targets, the creation of personalized medicine based on omics markers, and the discovery of connections between drugs and illnesses.

When it comes to suggesting powerful medication ideas and correctly anticipating both their qualities and potential toxicity hazards, DL has shown exceptional effectiveness. The analysis of enormous datasets, arduous compound screening while minimizing standard error, and the requirement for major R&D costs and time of over US$ 2.5 billion each decade may all be avoided with the application of AI approaches. With the aid of AI technology, new research may be conducted to aid in the discovery of new drug targets, logical medication design, and drug repurposing [44].

AI-based therapeutics in CRC

Chemotherapy, nCRT, and more comprehensive methods of treatment are available for CRC. Utilizing AI for CRC treatment, clinicians can choose the best-suited treatment option and increase the effectiveness of treatment by creating a personalized treatment course for each patient [1, 48]. AI-based interventions have been proven a state-of-the-art method to identify the appropriate surgery method, especially in handling complicated situations in CRC patients [49]. Further, these methods have been proven to be indispensable tools in the investigation of the precise stage of heterogeneity level of CRC during its diagnostic and suggest the possible management method. AI and ML present the ability to achieve early detection and diagnosis by precisely detecting polyps and lesions through image analysis. AI plays a promising role in improving accuracy and efficiency, especially in image analysis and molecular profiling [3]. ML identifies CRC biomarkers for non-invasive screening, while neural networks assist in analyzing the histopathologic images and reduce the expertise gaps. AI boosts medical image readability and guides precise robotic surgery, thus benefiting CRC treatment. AI also enhances nCRT, improving CRC treatment and efficacy assessment [50]. The table offers details about research on AI models for CRC treatment in relation to chemotherapy and nCRT (Table 2).

Table 2

Recent research on AI models for predicting nCRT and chemotherapy response in the treatment of CRC

Topic	Research	Model	Performance	Year	Reference
nCRT	EUS images of 43 LARC patients as predictive biomarkers Images pre-processed by lee, wiener, median, frost, bilateral, and wavelet filters	LR and SVM	AUC: 0.71 and 0.76 Accuracy: 70.0% and 71.5% Sensitivity: 69.8% and 80.2% (respectively)	2022	[51]
	CT images of 215 LARC patients Images evaluated by filtration histogram texture analysis and fractal dimension	LR	Accuracy: 82% Specificity: 89% Sensitivity: 60%	2021	[52]
	pCR prediction in 282 LARC patients (248 training and 34 validation)	ANN	AUC/accuracy/sensitivity: 0.84/0.88/0.94 respectively	2020	[53]
	pCR prediction in 6,555 non-metastatic cancer patients undergoing radical resection	LR	92.4%/88.2%: With/without—pathological complete response (overall survival rate of 3 years)	2019	[54]
	MRI of 98 patients (53/45: training test/validation set respectively) Image preprocessing by EMLMs and LOG filters	SVM, NN, BN, and KNN	Test (AUC and accuracy): 97.8% and 92.8% Validation (AUC and accuracy): 95% and 90%	2019	[55]
	MRI of 55 LARC patients to predict pCR and pNR rates	RF	0.83: Mean of AUC	2019	[56]
Chemotherapy	Irinotecan drug toxicity prediction in 20 metastatic CRC patients (liver function bloody tests and tumor markers)	SVM	Accuracy: 91%/76%/75% for diarrhea/leukopenia/neutropenia respectively	2019	[57]
Chemotherapy	Detection of IC₅₀ of a drug Evaluation of QSAR using NMR Analysis of 18,850 organic compounds	KNN, RF, and SVM	Above 63% accuracy	2018	[58]

AUC: area under curve; NN: neural network; BN: Bayesian network; LARC: locally advanced rectal cancer; EUS: endorectal ultrasound; IC₅₀: half maximal inhibitory concentration; LOG: Laplacian of Gaussian; NMR: nuclear magnetic resonance; CT: computed tomography; MRI: magnetic resonance imaging; pCR: pathologic complete response; EMLMs: ensemble machine learning models; pNR: pathologic non responder

The following table enlists the FDA-approved individual drugs for CRC treatment. However, the drug combinations are not FDA-approved but the drugs individually are approved by FDA. The list does not include all the drugs and there may be more drugs (Table 3) [59].

Table 3

FDA-approved medications for colon and rectal cancer consisting of both generic and brand names

FDA-approved individual drugs (CRC)	Mechanism of action
Ipilimumab	Prevent inhibition of T-cell mediated immune responses to tumors by binding CTLA-4
Bevacizumab (Mvasi)	Mvasi, a mAb, inactivates serum VEGF treating metastatic CRC
Pembrolizumab	Binds to PD-1 inhibiting its interaction with PD-L1 and PD-L2. Enhances anti-tumor immune response and tumor immune monitoring
Tucatinib	Inhibits HER-2 suppressing tumors by affecting cell proliferation and AKT and MAPK signaling
Ziv-aflibercept (Zaltrap)	Inhibits VEGF-A and PIGF to reduce vascular permeability and inhibit neovascularization. Delay vision loss and advancement of metastatic CRC
Irinotecan hydrochloride	Restricts DNA strands from elegating by attaching to the topoisomerase I-DNA complex and inhibiting its action, which causes fatal double-stranded splits in the DNA. Causes apoptosis as DNA damage is ineffectively repaired
Nivolumab	Restores tumor-specific T-cell response in patients. Binds to PD-1, avert PD-L1 and PD-L2 from blocking the action of T-cells
Ramucirumab	By binding to VEGFR2, it halts the ligands (VEGF-A, VEGF-C, and VEGF-D) from binding to it, blocking VEGF-stimulated receptor phosphorylation and the proliferation, permeability, and migration of human endothelial cells that are later caused by ligands
Cetuximab	Blocks the binding of EGF, which in turn prevents EGFR activation. It also binds selectively to the EGFR and phosphorylates and activates receptor-associated kinases (MAPK, PI3K/AKT, and JAK/STAT)
Regorafenib	At clinically attained concentrations it inhibits the function of VEGFR1, VEGFR2, VEGFR3, RET, PDGFR-alpha, PDGFR-beta, KIT, FGFR1, FGFR2, BRAFV600E, PTK5, TIE2, TrkA, RAF-1, BRAF, DDR2, SAPK2, Eph2A, and Abl
Panitumumab	Panitumumab binds primarily to EGFR and prevents ligands from binding to EGFR
Leucovorin calcium	By strengthening the bond between the active metabolite (5-FdUMP) and the enzyme thymidylate synthetase, leucovorin increases the action of fluorouracil

CTLA-4: cytotoxic T lymphocyte antigen-4; mAb: monoclonal antibody; PD-1: programmed death-1; PD-L1: PD-ligand-1; HER-2: human epidermal growth factor receptor-2; AKT: AKT serine/threonine kinase; MAPK: mitogen-activated protein kinase; VEGF-A: vascular endothelial growth factor-A; PIGF: placental growth factor; I-DNA: I-motif DNA; VEGFR2: VEGF receptor 2 ; EGF: epidermal growth factor; EGFR: EGF receptor; PI3K: phosphatidylinositol-4,5-bisphosphate 3-kinase; JAK: Janus kinase; STAT: signal transducer and activator of transcription; RET: Ret proto-oncogene; PDGFR-alpha: platelet derived growth factor receptor alpha; KIT: KIT proto-oncogene, receptor tyrosine kinase; FGFR1: fibroblast growth factor receptor 1; PTK5: protein tyrosine kinase 5; TIE2: tyrosine kinase with immunoglobulin like and EGF like domains 2; RAF-1: Raf-1 proto-oncogene, serine/threonine kinase; BRAF: B-Raf proto-oncogene, serine/threonine kinase; DDR2: discoidin domain receptor tyrosine kinase 2; SAPK2: sucrose non-fermenting-1-related protein kinase 2; Eph2A: ephrin type-A receptor 2; Abl: a gene; 5-FdUMP: 5-fluorodeoxyuridine monophosphate; BRAFV600E: B-Raf proto-oncogene; TrkA: tropomyosin receptor kinase A

The names of the drugs are shorthand references to the NCI’s cancer drug information. It’s possible that some drugs used to treat rectal and colon cancer aren’t included here. Some drugs combination used currently for CRC viz capecitabin and oxaliplatin (CAPOX), leucovorin, fluorouracil, and irinotecan (FOLFIRI), FOLFIRI-CETUXIMAB, oxaliplatin, 5-fluorouracil, and folinic acid (FOLFOX), capecitabine and irinotecan (XELIRI), 5-fluorouracil-leucovorin (FU-LV), FOLFIRI-BEVACIZUMAB, and oxaliplatin and capecitabine (XELOX) [59].

Challenges

Certain components of the drug development process haven’t gained enough attention yet. For instance, it is currently challenging to determine precisely how well a drug candidate binds to its intended protein target [24, 60, 61]. AI and other computational techniques do not currently perform well in this field for several reasons [24, 43, 62].

First off, AI is a data-mining technique. When using AI for data mining, the amount and quality of the available data directly affect how well AI models work [20, 24, 63–65]. Large volumes of training data are necessary for effective DNN training. The creation of transfer learning technology, that applies the lessons it picks up from one activity to another, could be a viable solution to this issue. The second problem is that occasionally the data quality is not good enough for effective AI learning. Biological assays, techniques, or conditions are frequently different from those used to measure experimental data in public databases. A substance can provide completely different results from measurements made using several techniques, which are incomparable. Also, outstanding data could be found in public databases [66].

There have been many unresolved questions such as how AI can be utilized to reliably estimate the binding affinity of a novel drug considering that the scaffold is distinct from the training datasets available. How AI can predict changes in protein structure that can occur at microseconds or even second timeframes? If AI can predict a new drug’s complex physical properties, such as its capacity to pass through the brain-blood barrier (BBB), membrane permeability, etc.? The most important therapeutic targets in drug research would be revealed if AI could predict new G protein-coupled receptors (GPCR) allosteric sites [24]. Hence, selecting high-quality data from the raw inputs is a crucial step before doing specific AI operations. AI could provide the answer by automating data entry as well. Finally, when a 2D interpretation of 3D atomic space occurs for AI computations, crucial 3D target structural information is lost, including the chemical surrounding of the target protein’s ligand binding site, the drug molecule’s conformation, and the protein’s flexibility. Alternatively, proteins and drug molecules could be sampled in varied conformations and states within physiological settings using molecular dynamics (MD) simulations. The effectiveness of this method was recently demonstrated in a study that used AI and MD simulations to examine the ligand specificity of GPCRs. Shortly, it’s possible to get beyond the constraints of binding-affinity forecasts and other molecular property predictions by transferring data from MD to AI [24].

The fact that the role of DL techniques is a “dark secret” or “black box” must be emphasized [67]. During the training stage, a neural network is only given one particular input with a label. Even the person who created the network may not be aware of what is being examined at the intermediate phases or the reasoning behind the model’s conclusions because the features are not explicitly described. To sum up, a lot of effort has been put towards incorporating AI techniques to speed up the drug discovery and development process, but more effective applications of these techniques will be required before the complete potential of AI in drug discovery and development is achieved [24].

Conclusions

With the overwhelming increase of clinical data and advancements in ML techniques and especially DL techniques, AI has enhanced the potential in various clinical aspects of CRC. AI algorithms are used for CRC including CRC identification, therapeutic evaluation, survival prediction, etc. However, there is not much literature available on the application of AI in CRC treatment. For better results, data quantity and quality are the important factors to be improved for precise treatment. The rationale behind any DL algorithm’s conclusions is the accurate calculation of the binding affinity of a novel drug candidate, along with the type of treatment to be selected for and individuals whom AI has to advance with preciseness. Despite ground-breaking advancements in AI-infused medication design and research, there is still a long way to go before personalized therapy for cancer patients can be effectively applied. This demonstrates the potential of AI technology along with current limitations.

Abbreviations

three-dimensional

artificial intelligence

ANN

artificial neural network

CNNs

convolutional neural networks

CRC

colorectal cancer

drug designing

deep learning

Eqn. 1

equation 1

FDA

Food and Drug Administration

FOLFIRI

leucovorin, fluorouracil, and irinotecan

logistic regression

molecular dynamics

machine learning

nCRT

neoadjuvant chemoradiotherapy

R&D

research and development

random forest

RNNs

recurrent neural networks

SVMs

support vector machines

Declarations Acknowledgments

We are thankful to Department of Biotechnology, Jaypee Institute of Information Technology, Noida for necessary facilities.

Author contributions

KD, PKT, MP, and CKJ: Conceptualization. KD and PKT: Investigation, Writing—original draft. KD, PKT, and MP: Visualization. SK, CKJ, RK, and SV: Writing—review & editing, Supervision. All authors read and approved the submitted version.

Conflicts of interest

The authors declare that they have no conflicts of interest.

Ethical approval

Not applicable.

Consent to participate

Not applicable.

Consent to publication

Not applicable.

Availability of data and materials

Not applicable.

Funding

Not applicable.

Qiu

Ding

Liu

Wang

Applications of artificial intelligence in screening, diagnosis, treatment, and prognosis of colorectal cancer

Curr Oncol 2022 29 1773 95

10.3390/curroncol29030146

35323346

PMC8947571

Miller

Nogueira

Devasia

Mariotto

Yabroff

Jemal

et al.

Cancer treatment and survivorship statistics, 2022

CA Cancer J Clin 2022 72 409 36

10.3322/caac.21731

35736631

Mitsala

Tsalikidis

Pitiakoudis

Simopoulos

Tsaroucha

Artificial intelligence in colorectal cancer screening, diagnosis and treatment. A new era

Curr Oncol 2021 28 1581 607

10.3390/curroncol28030149

33922402

PMC8161764

Dekker

Tanis

Vleugels

JLA

Kasi

Wallace

Colorectal cancer

Lancet 2019 394 1467 80

10.1016/S0140-6736(19)32319-0

31631858

Stanzione

Verde

Romeo

Boccadifuoco

Mainenti

Maurea

Radiomics and machine learning applications in rectal cancer: current update and future perspectives

World J Gastroenterol 2021 27 5306 21

10.3748/wjg.v27.i32.5306

34539134

PMC8409167

Boniolo

Dorigatti

Ohnmacht

Saur

Schubert

Menden

Artificial intelligence in early drug discovery enabling precision medicine

Expert Opin Drug Discov 2021 16 991 1007

10.1080/17460441.2021.1918096

34075855

Helwig

The role of AI technology in prediction, diagnosis and treatment of colorectal cancer

Artif Intell Rev 2022 55 323 43

10.1007/s10462-021-10034-y

34248245

PMC8255052

Nowak-Sliwinska

Scapozza

Ruiz

i Altaba A

Drug repurposing in oncology: compounds, pathways, phenotypes and computational approaches for colorectal cancer

Biochim Biophys Acta Rev Cancer 2019 1871 434 54

10.1016/j.bbcan.2019.04.005

31034926

PMC6528778

Carracedo-Reboredo

Liñares-Blanco

Rodríguez-Fernández

Cedrón

Novoa

Carballal

et al.

A review on machine learning approaches and trends in drug discovery

Comput Struct Biotechnol J 2021 19 4538 58

10.1016/j.csbj.2021.08.011

34471498

PMC8387781

Sharma

Jain

Chatterjee

Applications of machine learning algorithms in cancer diagnosis

Saxena

Chandra

Artificial intelligence and machine learning in healthcare Singapore

Springer

2021

Patel

Shukla

Huang

Ussery

Wang

Machine learning methods in drug discovery

Molecules 2020 25

5277

10.3390/molecules25225277

33198233

PMC7696134

Sah

Machine learning: a review of learning types

BioRxiv 2020070230 [Preprint]. 2020 [cited 2023 Jun 10]. Available from: https://www.preprints.org/manuscript/202007.0230/v1

Vamathevan

Clark

Czodrowski

Dunham

Ferran

Lee

et al.

Applications of machine learning in drug discovery and development

Nat Rev Drug Discov 2019 18 463 77

10.1038/s41573-019-0024-5

30976107

PMC6552674

Zararsiz

Elmali

Ozturk

Bagging support vector machines for leukemia classification

IJCSI 2012 9 365 8

Heikamp

Bajorath

Support vector machines for drug discovery

Expert Opin Drug Discov 2014 9 93 104

10.1517/17460441.2014.866943

24304044

Salmi

Rustam

Naïve bayes classifier models for predicting the colon cancer. 9th Annual Basic Science International Conference (BaSIC 2019); 2019 Mar 20–21; Malang, Indonesia. Bristol: IOP Publishing; 2019. pp. 1–8.

Saxena

Chandra

Artificial intelligence and machine learning in healthcare Singapore

Springer

2021

Shrestha

Mahmood

Review of deep learning algorithms and architectures

IEEE 2019 7 53040 65

10.1109/ACCESS.2019.2912200

Pouyanfar

Sadiq

Yan

Tian

Tao

Reyes

et al.

A survey on deep learning: algorithms, techniques, and applications

ACM Comput Surv 2018 51 1 36

10.1145/3234150

LeCun

Bengio

Hinton

Deep learning

Nature 2015 521 436 44

10.1038/nature14539

26017442

Gupta

Srivastava

Sahu

Tiwari

Ambasta

Kumar

Artificial intelligence to deep learning: machine intelligence approach for drug discovery

Mol Divers 2021 25 1315 60

10.1007/s11030-021-10217-3

33844136

PMC8040371

Cui

Aouidate

Wang

Yuan

Discovering anti-cancer drugs via computational methods

Front Pharmacol 2020 11

733

10.3389/fphar.2020.00733

32508653

PMC7251168

Hinkson

Madej

Stahlberg

Accelerating therapeutics for opportunities in medicine: a paradigm shift in drug discovery

Front Pharmacol 2020 11

770

10.3389/fphar.2020.00770

32694991

PMC7339658

Chan

HCS

Shan

Dahoun

Vogel

Yuan

Advancing drug discovery via artificial intelligence

Trends Pharmacol Sci 2019 40 592 604 Erratum in: Trends Pharmacol Sci. 2019;40:801.

10.1016/j.tips.2019.06.004

31320117

Karger

Kureljusic

Using artificial intelligence for drug discovery: a bibliometric study and future research agenda

Pharmaceuticals (Basel) 2022 15

1492

10.3390/ph15121492

36558943

PMC9785219

Farghali

Kutinová

Canová N

Arora

The potential applications of artificial intelligence in drug discovery and development

Physiol Res 2021 70 S715 22

10.33549/physiolres.934765

35199553

PMC9054182

Zhavoronkov

Vanhaelen

Oprea

Will artificial intelligence for drug discovery impact clinical pharmacology?

Clin Pharmacol Ther 2020 107 780 5

10.1002/cpt.1795

31957003

PMC7158211

Ramsundar

Eastman

Walters

Pande

Deep learning for the life sciences: applying deep learning to genomics, microscopy, drug discovery, and more 1st ed

Sebastopol (CA)

O’Reilly Media

2019

Nag

Baidya

ATK

Mandal

Mathew

Das

Devi

et al.

Deep learning tools for advancing drug discovery and development

3 Biotech 2022 12

110

10.1007/s13205-022-03165-8

35433167

PMC8994527

ORGANIC [Internet]

San Francisco (CA): GitHub; c2023 [cited 2023 Jun 10]. Available from: https://github.com/aspuru-guzik-group/ORGANIC

Steiner

Wolf

Glatzel

Andreou

Granda

Keenan

et al.

Organic synthesis in a modular robotic system driven by a chemical programming language

Science 2019 363

eaav2211

10.1126/science.aav2211

30498165

Wang

Zhang

Improving scoring-docking-screening powers of protein-ligand scoring functions using random forest

J Comput Chem 2017 38 169 77

10.1002/jcc.24667

27859414

PMC5140681

Wan

Zhu

Dai

Cai

Chen

et al.

DeepCPI: a deep learning-based framework for large-scale in silico drug screening

Genomics Proteomics Bioinformatics 2019 17 478 95

10.1016/j.gpb.2019.04.003

32035227

PMC7056933

Feinberg

Sur

Husic

Mai

et al.

PotentialNet for molecular property prediction

ACS Cent Sci 2018 4 1520 30

10.1021/acscentsci.8b00507

PMC6276035

Liaw

Sheridan

Svetnik

Demystifying multitask deep neural networks for quantitative structure-activity relationships

J Chem Inf Model 2017 57 2490 504

10.1021/acs.jcim.7b00087

28872869

Stork

Chen

Šícho

Kirchmair

Hit Dexter 2.0: machine-learning models for the prediction of frequent hitters

J Chem Inf Model 2019 59 1030 43

10.1021/acs.jcim.8b00677

30624935

Mayr

Klambauer

Unterthiner

Hochreiter

DeepTox: toxicity prediction using deep learning

Front Environ Sci 2016 3

00080

10.3389/fenvs.2015.00080

Awale

Reymond

Polypharmacology Browser PPB2: target prediction combining nearest neighbors with machine learning

J Chem Inf Model 2019 59 10 7

10.1021/acs.jcim.8b00524

30558418

Coley

Rogers

Green

Jensen

SCScore: synthetic complexity learned from a reaction corpus

J Chem Inf Model 2018 58 252 61

10.1021/acs.jcim.7b00622

29309147

Durrant

McCammon

NNScore 2.0: a neural-network receptor-ligand scoring function

J Chem Inf Model 2011 51 2897 903

10.1021/ci2003889

22017367

PMC3225089

Yasuo

Sekijima

Improved method of structure-based virtual screening via interaction-energy-based learning

J Chem Inf Model 2019 59 1050 61

10.1021/acs.jcim.8b00673

30808172

Olivecrona

Blaschke

Engkvist

Chen

Molecular de-novo design through deep reinforcement learning

J Cheminform 2017 9

10.1186/s13321-017-0235-x

29086083

PMC5583141

Jiménez-Luna

Grisoni

Weskamp

Schneider

Artificial intelligence in drug discovery: recent advances and future perspectives

Expert Opin Drug Discov 2021 16 949 59

10.1080/17460441.2021.1909567

33779453

Mak

Pichika

Artificial intelligence in drug development: present status and future prospects

Drug Discov Today 2019 24 773 80

10.1016/j.drudis.2018.11.014

30472429

Schneider

Walters

Plowright

Sieroka

Listgarten

Goodnow

RA Jr

et al.

Rethinking drug design in the artificial intelligence era

Nat Rev Drug Discov 2020 19 353 64

10.1038/s41573-019-0050-3

31801986

Zhang

Tan

Han

Zhu

From machine learning to deep learning: progress in machine intelligence for rational drug discovery

Drug Discov Today 2017 22 1680 5

10.1016/j.drudis.2017.08.010

28881183

Chen

Liu

Zhang

Chen

Artificial intelligence for drug discovery: resources, methods, and applications

Mol Ther Nucleic Acids 2023 31 691 702

10.1016/j.omtn.2023.02.019

36923950

PMC10009646

Bondeven

Laurberg

Hagemann-Madsen

Ginnerup

Pedersen B

Suboptimal surgery and omission of neoadjuvant therapy for upper rectal cancer is associated with a high risk of local recurrence

Colorectal Dis 2015 17 216 24

10.1111/codi.12869

25510260

Quero

Mascagni

Kolbinger

Fiorillo

Sio D

Longo

et al.

Artificial intelligence in colorectal cancer surgery: present and future perspectives

Cancers (Basel) 2022 14

3803

10.3390/cancers14153803

35954466

PMC9367568

Yin

Yao

Zhang

Application of artificial intelligence in diagnosis and treatment of colorectal cancer: a novel prospect

Front Med (Lausanne) 2023 10

1128084

10.3389/fmed.2023.1128084

36968824

PMC10030915

Abbaspour

Abdollahi

Arabalibeik

Barahman

Arefpour

Fadavi

et al.

Endorectal ultrasound radiomics in locally advanced rectal cancer patients: despeckling and radiotherapy response prediction using machine learning

Abdom Radiol (NY) 2022 47 3645 59

10.1007/s00261-022-03625-y

35951085

Tochigi

Kamran

Parakh

Noda

Ganeshan

Blaszkowsky

et al.

Response prediction of neoadjuvant chemoradiation therapy in locally advanced rectal cancer using CT-based fractal dimension analysis

Eur Radiol 2022 32 2426 36

10.1007/s00330-021-08303-z

34643781

Huang

Tsai

Chang

et al.

Machine learning for predicting pathological complete response in patients with locally advanced rectal cancer after neoadjuvant chemoradiotherapy

Sci Rep 2020 10

12555

10.1038/s41598-020-69345-9

32724164

PMC7387337

Tan

Kong

Jiang

Chen

et al.

Predictors and risk factors of pathologic complete response following neoadjuvant chemoradiotherapy for rectal cancer: a population-based analysis

Front Oncol 2019 9

497

10.3389/fonc.2019.00497

31263674

PMC6585388

Shayesteh

Alikhassi

Fard

Esfahani A

Miraie

Geramifar

Bitarafan-Rajabi

et al.

Neo-adjuvant chemoradiotherapy response prediction using MRI based ensemble learning method in rectal cancer patients

Phys Med 2019 62 111 9

10.1016/j.ejmp.2019.03.013

31153390

Ferrari

Mancini-Terracciano

Voena

Rengo

Zerunian

Ciardiello

et al.

MR-based artificial intelligence model to assess response to therapy in locally advanced rectal cancer

Eur J Radiol 2019 118 1 9

10.1016/j.ejrad.2019.06.013

31439226

Oyaga-Iriarte

Insausti

Sayar

Aldaz

Prediction of irinotecan toxicity in metastatic colorectal cancer patients based on machine learning models with pharmacokinetic parameters

J Pharmacol Sci 2019 140 20 5

10.1016/j.jphs.2019.03.004

31105026

Cruz

Gomes

Borralho

Rodrigues

CMP

Gaudêncio

Pereira

In silico HCT116 human colon cancer cell-based models en route to the discovery of lead-like anticancer drugs

Biomolecules 2018 8

10.3390/biom8030056

30018273

PMC6164384

Drugs approved for colon and rectal cancer [Internet]

Maryland: National Cancer Institute; c2023 [cited 2023 Jun 6]. Available from: https://www.cancer.gov/about-cancer/treatment/drugs/colorectal

Clark

Negron

Hauser

Sun

Wang

Abel

et al.

Relative binding affinity prediction of charge-changing sequence mutations with FEP in protein-protein interfaces

J Mol Biol 2019 431 1481 93

10.1016/j.jmb.2019.02.003

30776430

PMC6453258

Das

Krein

Breneman

Binding affinity prediction with property-encoded shape distribution signatures

J Chem Inf Model 2010 50 298 308

10.1021/ci9004139

20095526

PMC2846646

Śledź

Caflisch

Protein structure-based drug design: from docking to molecular dynamics

Curr Opin Struct Biol 2018 48 93 102

10.1016/j.sbi.2017.10.010

29149726

Blomme

Will

Toxicology strategies for drug discovery: present and future

Chem Res Toxicol 2016 29 473 504

10.1021/acs.chemrestox.5b00407

26588328

Keefer

Chang

Kauffman

Extraction of tacit knowledge from large ADME data sets via pairwise analysis

Bioorg Med Chem 2011 19 3739 49

10.1016/j.bmc.2011.05.003

21616672

Zhong

Xing

Liu

Xiong

et al.

Artificial intelligence in drug design

Sci China Life Sci 2018 61 1191 204

10.1007/s11427-018-9342-2

30054833

Ding

Gudivada

Augmentation and evaluation of training data for deep learning. 2017 IEEE International Conference on Big Data (Big Data); 2017 Dec 11-14; Boston (MA), USA. New York (NY): IEEE. pp. 2603–11.

10.1109/BigData.2017.8258220

Voosen

The AI detectives

Science 2017 357 22 7

10.1126/science.357.6346.22

28684483