Genetics of Alzheimer's Disease - Paper 1

Alzheimer disease is the most common cause of dementia in the elderly individuals. Clinically, patients show presence of symptoms that include short-term memory loss, executive dysfunction, confusion, agitation, and behavioral disturbances. Three have been associated with autosomal dominant familial AD (APP, PSEN1, and PSEN2) and 1 genetic risk factor (APOEε4 allele). The disease is divided into 2 subtypes based on the age of onset: early-onset AD (EOAD) and late-onset AD (LOAD). This review provides an overview of the clinical and genetic features of AD.

Genes associated with autosomal dominant AD include AD1: Amyloid precursor protein (APP), AD3: Presenilin 1 (PSEN1) and AD4: Presenilin 2 (PSEN2), while gene associated with Sporadic AD is AD2: APOE. Various aspects of the genes include, Inheritance and clinical features, gene location and structure, Gene function and structure, and gene variation.

AD1:Amyloid precursor protein (APP)

APP is a type I integral membrane glycoprotein, from which Aβ gets derived by enzymatic cleavage. It was initially purified from plaque and vascular amyloid deposits. It is also seen in down syndrome (trisomy 21) patients who developed amyloid deposits, and neuropathological features of AD.

APP gene is spliced into several products based on their length in amino acid. It is expressed by tissue types with 3 isoforms relevant to AD and are restricted to the central nervous system (APP695) or expressed in both the peripheral and central nervous system (CNS) tissues (APP751 and APP770).

Amyloid precursor protein function is implicated in neural plasticity and as a regulator of synapse formation, though its function is unknown. APP and its by-product Aβ also have been found to be translocated inside mitochondria and implicated in mitochondrial dysfunction.

APP transcripts are identified in exons 7,8 and 15 are alternatively spliced. Variants include APP695 contains exon 15 but excludes exons 7 and 8; APP751 and APP770 contain exons 16 and 17 but exclude exons 7, 8, and 15; APP770 splice variant83, producing the Aβ peptide is located within exons 16 and 17. L-APPs splice variants are observed with missing exon 15 and in various combinations with exons 7 and 8. Alternative splicing of exons 7 and 8 is modulated in the brain during aging and possibly during AD. Though most of the splice variants contain Aβ-encoding sequences, 2 additional rare transcripts, APP365 and APP563, do not implicate variability in APP isoform function.

AD3: Presenilin 1 (PSEN1)

Common cause of EOFAD is the mutations in PSEN1 or presenilin 1 gene. PSEN1 are the major components of the atypical aspartyl protease complexes, located as AD3 locus on chromosome 14 with positional cloning. PSEN1 associated AD shows clinical symptoms of progressive dementia, parkinsonism, notch signaling modulation. and Aβ intracellular domain generation. Further Spastic paraparesis is found in some cases along with neuronal loss, abundant Aβ neuritic plaques and neurofibrillary tangles, and degeneration extending into the brain stem.

PSEN1 is located on chromosome 14q24.2 consisting of 12 exons that encode a 467-amino-acid protein.

PSEN1 is expressed as a polytopic membrane protein that forms the catalytic core of the γ-secretase complex with γ-secretase found at the cell surface, Golgi, endoplasmic reticulum, and in mitochondria. Main function of PSEN1 is its involvement in cognitive memory.

Variations in PSEN1 are missense mutations causing aminoacid substitutions throughout the PSEN1 protein resulting in an increased ratio of Aβ42 to Aβ40 peptides.

AD4: Presenilin 2 (PSEN2)

Presenilin2 (PSEN2) is another gene for the chromosome 1 AD4 locus, with high homology to the AD3 locus(PSEN1). Missense mutations of PSEN2 gene rarely cause EOFAD. Age of onset is highly variable among PSEN2-affected members of the same family.

The PSEN2 gene is located on chromosome 1 (1q42.13), identified by sequence homology and then cloned, with 12 exons, organized into 10 translated exons that encode a 448-amino-acid peptide. Structures depict 9 transmembrane domains and a large loop structure between the sixth and seventh domains, displaying tissue-specific alternative splicing.

Component of atypical aspartyl protease called γ-secretase which is responsible for the cleavage of Aβ is expressed in a variety of tissues, most primarily in neurons, is the gene PSEN2. Though functions are not explicit, differential expression of presenilin isoforms may lead to differential regulation of the proteolytic processing of the APP.

First mutation of the PSEN2 gene is a point mutation located within the second transmembrane. It results in the substitution of an isoleucine for an asparagine at residues 141 (N141l). The recent one, as mentioned in the paper, is the V393M mutation, located within the seventh transmembrane domain with a total of 14 PSEN2 mutations till date.

AD2: APOE

APOE gene is inherited in both familial late-onset and sporadic late-onset AD across multiple ethnic groups. The APOE ε4 genotype expressed in earlier age of onset in both AD and down syndrome, a worse outcome after head trauma and stroke that occurs both in humans and transgenic mice.

The APOE gene is located on chromosome 19 (19q13.2) consisting of 4 exons that encode a 299-amino-acid protein. Structure is in a cluster with other apolipoprotein genes: APOC1, APOC2, and APOC4 with the APOE ε4 loci are located within exon 4 of the gene. The 3 APOE ε4 alleles (ε2, ε3, and ε4) are defined by 2 single nucleotide polymorphisms, rs429358 and rs7412, which encode 3 protein isoforms (E2, E3, and E4).

APOE function includes distribution and metabolism of cholesterol and triglycerides within many organs and cell types in the human body. It enhances the deposition of the Aβ peptide, and isoforms may modulate the amount of lipid available for neurons.

Gene dose of APOE ε4 is a major risk factor for AD. Reports an association between gene dose, age at onset and cognitive decline. Different variations include APOE 2/2, 3/3, 2/3, 3/4, 4/4 genotypes.

To conclude, Alzheimer disease can be characterized as an irreversible, progressive loss of memory and decline in cognitive skills, with no cure, and treatments only manage to slow down the progression of the disease. With biology behind AD remaining unclear, AD is termed as a genetic and environmentally complex disease.

Link to the paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3044597/

Early Stage Alzheimer's Disease: Getting Trial Ready - Paper 3

Most important healthcare challenge leading to a progressive and fatal disease associated with age is Alzheimer's, which emphasizes therapeutic intervention at early stages of the disease. Treatments targeting the core pathologies of AD mostly yield disappointing results in clinical trials as the intervention is too late in the course of the disease.

Research paper reviews on trial-ready cohorts, which are deemed effective for identifying participants eligible for clinical trials in early-stage Alzheimer's. It is also important that building these cohorts concerning AD, involves characterizing individuals with cognitive testing and assessing key biomarkers such as PET or CSF measures of amyloid pathology. To minimize participant burden, existing trial-ready cohort programs have leveraged online tools that allow web-based recruitment and data collection. The Paper also explains how progress can be made towards efficient trials in early-stage AD, by examining and learning from the approaches used to establish AD trial-ready cohorts and highlights the future efforts on design and implementation.

Furthermore, paper details on building these cohorts efficiently and effectively that mainly lays on considerable planning and technological infrastructure to facilitate recruitment, remote longitudinal assessment, data management and data storage. Various clinical trials launched as listed in the paper includes TRC-PAD, The Trial-Ready Cohort for Preclinical and Prodromal Alzheimer's Disease funded by the NIA, National institute on aging; J-TRC, The Japanese Trial-Ready Cohort, representing an international collaboration between investigators at the University of Tokyo, Japan, and the TRC-PAD investigators; and EPAD Registry, The European Prevention of Alzheimer's Dementia, project of the Innovative Medicines Initiative.

Participants for Trial-Ready Cohort for Preclinical and Prodromal Alzheimer's Disease are recruited primarily from existing registries and remote consent to be followed in the Alzheimer Prevention Trials Web study (APT Web study) for quarterly unsupervised assessments. For in-person assessment, algorithms are run to select participants through the site referral system. Various Screening includes cognitive and clinical assessments and amyloid testing. Participants , who become eligible are enrolled for semi-annual clinical visits until when an appropriate clinical trial becomes available at their site. Older Japanese individuals were the participants of J-TRC to adapt the informatics and biostatistics approaches developed for TRC-PAD to build an exploratory cohort. Efforts were facilitated by using cloud-based technology. EPAD project worked on the basis of four aims: improving access to cohorts and registries, developing an EPAD Registry, establishing a longitudinal cohort and performing an adaptive proof-of-concept trial.

Paper further highlights that trial-ready cohorts also existed for genetically determined populations who are at risk of developing AD, such as those with familial AD and Down syndrome. The DIAN-TU platform was launched to accelerate the identification of effective drugs for treating and preventing ADAD, an autosomal dominant form of AD. Individuals with Down syndrome who are at very high risk of developing AD pathology, The Alzheimer's Biomarker Consortium-Down Syndrome (ABC-DS) Project was launched. It works with ACTC-DS, a network of 16 international clinical trial sites including all ABC-DS sites.

Considerations for highlighting these trial cohorts on global scale, different aspects that are listed as focal points in the paper include data capture and management approaches, regulatory requirements, representative recruitment, participant retention, use of optimal remote and unsupervised assessments, and the use of appropriate statistical models in risk assessment.

To conclude, Use of web-based programs, and statistical modeling including demographic data, subjective to cognitive symptoms, cognitive performance, genetic markers can be used as a selection point for individuals for biomarker testing. Pre-randomization data can facilitate the aims of trial-ready cohort programs. Transfer of biomarker data from trial screening visits into the TRC-PAD data system during pandemic is important. To obtain eventual neuropathological correlation with clinical and biomarker data, use of the trial-ready cohorts provide an opportunity and support in long-term safety and efficacy assessments.

Future directions could be plasma assays that measure Aβ peptides or phosphotau species, which can be used to detect early neurobiological changes of AD even prior to CSF and PET amyloid biomarker positivity. Candidates can be selected for primary prevention treatment by developing accurate metabolic profiling of amyloid dysregulation, providing a path to blood-based indicators. Large-scale prescreening of participants for primary prevention trials can be facilitated by remote plasma testing, and by using commercial clinical test sites to obtain, process, and ship specimens to central laboratories. Paper thereby provides a detailed review on how millions of people worldwide can be provided with effective treatments to slow the progression of AD with methods of streamlined referral to trial-ready cohorts and clinical trials.

Link to the paper: https://www.nature.com/articles/s41582-022-00645-6#:~:text=A%20trial%2Dready%20cohort%20is,CSF%20measures%20of%20amyloid%20pathology.

2021 marks a new Era for Alzheimer's Therapeutics - Paper 5

This review provides a brief outlook on the recent findings in AD research.

Detrimental effect on older populations and patients with dementia due to COVID-19 pandemic robusts the therapeutic approach to Alzheimer's disease research.

Aducanumab, the first-disease modifying therapy, targeting amyloid β is approved by the US Food & Drug Administration. This set as an example to develop further therapies. Outcome of this approval laid pathway to develop guidelines to identify patients for testing.

Phase 2 trial of Donenamb led to another therapeutic development in early ADs. Donenamb reduced amyloid β, setting it as a composite score for cognition, compared to placebo at 76 weeks.

Another therapy in the spotlight targeting amyloid β and tau, is the ADAMANT phase 2 study that assured a promising response of the tau vaccine AADvac1 in patients with mild AD. This therapeutic study appraised the vaccine's safety, tolerability, immunogenicity, clinical efficacy, and biomarker response.

Biomarkers behaved as a trendsetter in the approval process of aducanumab and in the other phase 2 studies as treatment efficacy and assessment of target engagement.

Molecular imaging biomarkers were pivotal in 2021, contributing to the understanding of disease mechanisms. Discovery on the role of genetic variants of APOE on tau pathology and neurodegeneration was remarkable. Multiconsortia throughout the world, started to rigorously work on the AD Neuroimaging Initiative's findings in Lewy body dementia, FTD and vascular cognitive impairment. Highlights were the biomarkers investigation in the prodromal and preclinical stages of these disorders with AD pathology leaning towards targeting the individualized therapies that address the proteinopathies.

Link to the paper: https://www.thelancet.com/pdfs/journals/laneur/PIIS1474-4422(21)00412-9.pdf

Alzheimer's Disease: Experimental Models and Reality - Paper 7

Experimental models are essential to understand and perform preclinical testing for new therapies of AD. The vast majority comprises animal models, that includes transgenic mice, Drosophila melanogaster and Caenorhabditis elegans, and vertebrates such as zebrafish. Models were designed to develop amyloid plaques and FTs in their brains but often lack the presence of other pathological features that define AD, which include neuronal loss and most importantly, development of neurofibrillary tangles.

Transgenic mice are used in transgenic mouse models, because “wild-type mice APP” has 97% sequence homology compared with human APP. The sequence differences between humans and mice include 3 amino acids (R5G, Y10F and H13R), located within the Aβ sequence, impairing Aβ aggregation and preventing the formation of amyloid plaques in wild-type mice. AD-associated pathological features occurred in these models, resulting in consistent plaque pathology. These pathologies are dependent on the expression of FAD mutations, resulting in clinical trials on sAD patients, blocking the translatability of success in these models.

Transgenic mice expressing APP and PSEN1 with FAD mutations were commonly used in AD studies. Models expressed as the associations of APP with FAD mutation, were carried out by promoters, causing over-expressions of APP, resulting in human AD pathology. PDAPP mouse and Tg2576 mouse models were used, using Indiana and Swedish mutations respectively, with PDGF-β and PrP promoters, resulting in various formations and symptoms in the brain regions. Multiple FAD associated mutations in transgenic mouse models expressed severe pathology with early plaque formation, beginning at approximately 6 weeks up to 6 months. In the presence or absence of human PSEN1, they developed vigorous plaque formation in brain regions that are rich in areas of plaques found in AD patients, such as the cortex and hippocampus. Limitations of these transgenic models are the lack of neurodegeneration and regional brain atrophy that occurs in human AD.

Transgenic mice expressing tau are identified by sequence differences between mouse and human tau that express isoform, as seen in humans, which causes tangle formation in mice. Most commonly used “disease models” that express 4R ,with P301L or P301S mutations, are the Frontotemporal lobar degeneration(FTLD). FTLD is a clinically and pathologically heterogeneous syndrome, characterized by the progressive decline in behavior or language associated with degeneration of the frontal and anterior temporal lobes. In transgenic mice, NFTs form readily displaying human tau with mutations, when associated with FTLD, and these mice develop NFTs, neurodegeneration, atrophy and motor deficits. Limitations are the necessity of human tau mutations for the development of NFT, and developed mutated tau can cause toxicity or interaction with Aβ, which does not occur in case of AD. Significant motor deficits and interference with cognitive testing widely occurs with over-expression of mutated tau, that are not seen in AD.

Models are based on the concurrent expression of mutated forms of APP, MAPT and PSEN1 or PSEN2 resulting in the observation of transgenic mice expressing both plaques and tangles not until old age in these models. Most completed transgenic mouse model of AD pathology reported is the 3xTg mouse model. It was noticed that the 3xTg mice first developed intraneuronal Aβ at 3–4 months, followed by plaque development at approximately 6 months in the cortex and hippocampus, while NFTs formed initially in CA1 and later in the cortex at approximately 12 months. Mice exhibited localized neurodegeneration, synaptic impairment and cognitive deficits from 6 months. Downside is that the 3xTg mice are limited as the production of mutated Aβ and tau is not representative in sporadic AD (sAD), and is highly overexpressed. Presence of plaques and tangles cannot be seen, until the mice reach old age, and the pathology is less, as in AD.

Unique Transgenic models are developed in replicating the pathological feature of AD which was evident from the Tg-SwDI transgenic mouse model of CAA. It expresses Swedish (APP^K670N/M671L), Dutch (APP^E693Q) and Iowa (APP^D694N) APP FAD mutations. The Dutch and Iowa mutations are associated with the hereditary cerebral hemorrhage with amyloidosis (HCHWA),where there is extensive CAA with more limited plaque pathology. Testing methodologies are incorporated that help in reducing vascular amyloid deposits without complication, in an ongoing passive immunization AD clinical trial. Models with extensive CAA not inducing microhemorrhages are of critical importance. Another model, resulting in a unique phenotype of, showing significantly increased expression of Aβ oligomers, synaptic impairment and cognitive impairments, from 8 months of age, without the formation of plaque or tau pathology is the APP E693Δ-Tg model expressing the Osaka (APP^E693Δ) mutation. These two models - Tg-SwDI transgenic mouse model and APP E693Δ-Tg model are considered as the most beneficial as it replicates the pathological features of AD more vigorously than any other models. And the constraints are they cannot be used as a complete model of AD, as they do not replicate all the features of the disease.

Overlapping the other mouse models, Knock-in mouse model was developed by humanizing mouse Aβ and knocking in specific APP FAD mutations, to avoid the confounding effects of APP over-expression. In this model, the knock-in mice had the same expression of APP and AICD as wild-type mice and APP expression occurs in a physiological manner in the correct brain regions and cell types. Restrictions highlighted are, they are the models of FAD and not sAD, and the pathology only develops after “knock-in'' of a combination of specific multiple FAD mutations.

Transgenic rat models are potentially useful in AD research and offer specific advantages over transgenic mice, however their usage is limited as their suitability to AD is still to be determined.

Physiological models are considered better, as transgenic models curb to FAD and not sAD by explicitly displaying non-physiological pathology development. Physiological models are ideally considered models, as they represent changes that occur in sAD and multiple species develop neuropathological features similar to those seen in AD brain quite naturally.

Non-human primates have well characterized AD neuropathological features. This is because of their biological proximity to humans, behavioral complexity, large brains that are favorable for imaging studies or CSF collection, and with a natural accumulation of Aβ having 100% sequence homology with human Aβ. Great apes, like chimpanzees, gorillas and orangutans, and old world monkeys, like rhesus monkeys, cynomolgus monkeys, baboons and vervets were used in research studies. Studies on Aβ accumulation, development of amyloid plaques, cerebral amyloid angiopathy (CAA), and tauopathy development in their brains enable them to be used in AD studies. Mass spectrometry study reveals that squirrel monkeys have all major Aβ species that are present in the human brain. Gray mouse lemurs were also used in AD related preclinical trials. Non-human primates were deducted to have age related Aβ pathology, with tauopathy being rare and/or very limited. From the study, it was inferred that the rhesus monkey is the most practical nonhuman primate model to study AD as it is well characterized and the squirrel monkey is the best available non-human primate model to study CAA.

Other physiological models includes, Dogs and guinea pig relatives Octodon degu. These were characterized as naturally developing AD associated pathology species with age. Cognitive tests show that aged dogs can develop complex learning tasks, executive function, spatial learning and attention, and memory. Though aged dogs were used in preclinical therapeutic studies, limitations include lack of NFTs, lack of compact plaques, long lifespan and lack of consistent pathology. Octodon degu have high sequence homology with human Aβ and also discloses, intracellular and extracellular accumulation of Aβ, plaques at old ages, intracellular tau accumulation, astrocytosis, synaptic changes and memory impairment that correlates with increased levels of oligomers.

Physiological models have scientific and practical limitations that prevent widespread use of these models.

AD research studies use invertebrate animal models such as Drosophila, C. Elegans and lower order animal models such as zebrafish. Though Drosophila, C. Elegans and zebrafish express orthologues of some of the genes that are essential in AD pathology, the presence of these orthologues and the genetic similarity to human genes varies between species. Using such invertebrate models include easy handling, low cost and short life span of animals. Transgenic invertebrate models are used in high throughput genetic or drug screens as well in identifying modifiers of tau toxicity. Limitations of using such an invertebrate model is that the vast differences between humans and invertebrates shows lack of conserved functional pathways and the lack of important interactors/mediators involved in the downstream response of expressed human genes.

Factors to consider when choosing a model includes the model representing the new therapeutic approach on both plaques and tangles, model to address the issue of non-physiological over-expression of APP or tau, model to reason out the downstream pathological effects of the human protein expressed in transgenic mice, models that do not influence Aβ biochemistry and deposition in physiological models ,models representing sAD, AD transgenic models expressing human transgenes, and whether the model mediates cognitive impairment in transgenic models appear to be the same as the one that mediates impairment in humans.

AD being a uniquely human disease, requires careful examination of neuropathology and cognitive impairment in multiple species, including those closest to humans. Performing research using human tissue whenever possible, should be set as a standard to follow. Lack of translation between animal models and human studies has resulted in the development of more human-centric approaches. Main advantage of the animal model is that it provides the option to do preclinical testing, in vivo, allowing the testing of general toxicity of new therapeutics and providing a system in which cognitive testing can be done. Knock-in mouse models are considered as more representative and physiological models of AD. Genetic similarity to humans , physiological relevant development of pathology that is found in sAD makes non-human primates models to be considered better than transgenic models. However limitations of availability, costs, time until onset of phenotype and the inconsistent presence of pathology in all animals makes the model disadvantageous. Though human cell culture models have the advantage of allowing high-throughput screening of novel therapeutics by directly using human cells, these models cannot replace in vivo models for preclinical testing. To close out, the best approach is to perform preclinical testing in multiple animal models that exemplifies a unique aspect of AD pathology till a complete physiological animal model of sAD is available promising a greater translation of preclinical results to human clinical trials.

Link to the paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5253109/

Artificial Challenge for Alzheimer's Disease: Promise or Challenge? - Paper 9

Experimental and clinical research running for the past decades have untangled many mechanisms in the pathogenesis of Alzheimer's disease but the puzzle remains incomplete. Though the missing pieces exist, recent growth of data sharing initiatives , and the integrating Big data from multiple studies provides a map to explore the pathophysiological mechanisms of AD. Artificial Intelligence (AI) offers a wide variety of methods to analyze large and complex data to structure the AD field's working process. Paper focuses on recent findings and future challenges for AI in AD research to support clinical practices for predicting individual risk of AD to develop effective and personalized therapies. Paper also details the study of Computer-Aided Diagnosis tools for AD diagnosis.

Through continuous research methodologies, many mechanisms in the pathogenesis of the disease, such as the β-amyloid hypothesis, though have been revealed, many questions still remain unanswered. Need of handling voluminous data as forms of clinical and biological, from electronic health records and multi-omics sciences representing tens of thousands of AD patients outreaches the human ability to make sense of the disease. Unlimited information targets on biological processes, such as genomes, transcriptomes, and proteomes, which can be explored through Big Data exploitation.

Big Data can be extracted with meaningful information using complex AI-based models, however as the data becomes complex it becomes more strainful to interpret how they structure their output. That's the reason they are referred to as black-box models. Although AI technological development plays a pivotal role in healthcare applications in recent years, making AI explainable is a key problem. Both patients and clinicians need to trust research methods to make decisions about people's health. To benefit the most out of the biological experiments and refine the findings, healthcare professionals have to be assisted by complex biological modeling, based on mathematical and statistical tools such as Artificial Neural Networks. Novel analytic approaches from bioinformatics and statistics can be formalized with domain expertise from psychology, neuroscience, neurology, psychiatry, geriatric medicine, biology, and genetics can be applied on Big Data in AD research projects. This can be done using predictive models to address detailed questions about

promising biomarker combinations, patient subgroups, and disease progression, further leading to the development of effective treatment strategies, helping patients with tailored medical approaches.

Due to the potential for fast, low-cost, and accurate automation AI revolutionized the way digital data is analyzed and used. An extensive application of AI in the biomedical field is for Computer-Aided Diagnosis (CAD). With AI, upon data analysis the diagnostic process can be automated, leading to an early and differential diagnosis of AD or dementia of different etiology. Data derived from biologic, imaging, demographics, neuropsychological tests, electronic health records, and multi-omics are integrated and processed by several algorithms and tasks. Algorithms and tasks include Classification, Clustering and regression. Such tasks lead to useful outcomes for early and accurate diagnosis, prediction of the course and progression of disease, patient stratification, and discovery of novel therapeutic targets and disease-modifying therapies.

The fundamental branch of AI is Machine learning (ML), consisting of data analysis techniques that aim to generate predictive models by learning from data. Deep learning(DL) is a sub-field of ML. Methods are used to learn relationships between inputs and outputs by modeling highly non-linear interactions into higher representations. Other methods highlighted in the paper besides ML and DL include supervised learning, Unsupervised learning, Classification task, Regression task, Clustering, Overfitting, Ensemble learning, Transfer learning, Cox regression. Despite ample research effort, a cure capable of modifying and/or halting the course of the disease still does not exist.

AI implementation can be applied to analyze the literature and high-throughput compound screening data, to perform an initial molecular screening and automated chemical synthesis. AI can be used to propose a new molecular optimization plan and new bioassays can be conducted to evaluate the biological effects of the compound, to automate drug development cycles based on AI design and high-throughput bioassay. This technique helps accelerate new drug development, thereby repurposing known drugs for Alzheimer's. Analyzing large scale transcriptomics, molecular structure data and clinical databases, AI is used to predict drug repurposing which is a fast, low-cost drug development pathway. Using AI algorithms on genetic and clinical data, participant selection can be optimized and clinical trials can be simplified too. This can be done by coupling AI with data from wearables. Thus, AI has become a promising technology to support research and to develop novel effective therapies although only a few AI applications have made it to the clinical application stage.

During the omics era, when the technological advance in biochemistry, molecular biology and analytical chemistry enabled the development of genomics, transcriptomics, proteomics, metabolomics and metabonomics several databases have been established t. To specifically mention, the Gene Expression Omnibus (GEO), that collects functional genomics data of array and sequence based data regarding many physiological and pathological conditions, including AD.UK Biobank collects and stores healthcare databases and associated biological specimens for a wide range of health-related outcomes from a large prospective study including over 500,000 participants. For more comprehensive understanding of disease heterogeneity and personalized medicine and drug development, public and private databases represent the substrate and the source for AI.

Human capacity of data analysis of highly dimensional complex systems can be done mainly by ML algorithms. ML has been used in the CAD of many pathologies including AD. CAD tools functioning include Data collection, Data elaboration, Data analysis and Diagnosis. After data collection, data are elaborated to be made ready for the analysis using AI-based techniques. The outcome is available for diagnostic evaluation. By this way, both MRI and PET data can be modeled by ML algorithms, yielding accuracy. Integrating neuroimaging data with other variables, such as cognitive measures, genetic factors, or biochemical changes, enhances the model's performance.

A biomarker-based ML model for predicting AD-related cognitive decline can be established using autosomal-dominant AD data, including amyloid-PET, FDG-PET, structural MRI, and CSF. Model showed accurate prediction and successful generalization to an independent sample of sporadic AD patients in predicting cognitive decline. Using these findings, it can be inferred that biomarker-based ML can be efficiently used to derive meaningful prognostic indices to identify subjects at risk of imminent cognitive decline.

AD is a highly heterogeneous disease in clinical manifestations, disease progression, biological profiles, and response to pharmacological treatment. This complexity is a great issue for physicians in diagnosing the disease paving way to many failures in pharmacological trials. Clustering tools are used when a huge amount of multi-modal data relative to patients are available. These tools determine groups that share similar properties among populations. Each cluster defines a sub-phenotype with its own peculiar characteristics, providing fine-grained information for the stratification of patients. Clustering minimizes the intra-cluster distance, while maximizing the inter-cluster distance. Clinician assessment is pivotal in the development of individually-tailored treatments.

Data Integration from omics studies explores the pathophysiological mechanisms of AD. Deep Neural

Network models can be used to elaborate multiple heterogeneous omics data sets that involve gene expression and DNA methylation from prefrontal region tissue. These models represent tools to identify the biological samples typical to the distinguished AD patients. This application mainly helped identify five stable molecular subtypes of AD, that were validated with the RNA data and by bioinformatic analyses such as Gene Ontology

pathway enrichment and gene co-expression network analysis. These subtypes and their molecular signatures can guide the development of novel therapeutics for AD toward precision medicine.

Despite much recent progress, AI faces lots of challenges. Future perspective majorly drives in understanding the algorithms interpretation of datas and making further decisions. Explaining the decisional processes of black-box models is non-interpretable. AI has the potential to integrate infinite amounts of data across different modalities by increasing the performance of prediction thus stabilizing its usefulness in clinical practice. Negative impact can be due to continuous addition of different data types for multi-modal representation that joins irrelevant information. If the same is not done correctly, increasing the number of features gives more accurate predictions, indirectly impacting the model performance.

AI pipeline architecture is built based on the cross-industry standard process for data mining (CRISP-DM) framework, which identifies six major phases in data mining - business understanding, data understanding, data preparation, modeling, evaluation, and deployment. This substantiates the reason of the no free lunch (NFL) theorem that states no single machine learning algorithm is universally the best-performing algorithm for all

Problems. CRISP-DM-based pipelines require a lot of effort by human AI experts, requiring a continuous interaction between healthcare researchers and data scientists.

Culmination of AI would be the model to integrate heterogeneous data to improve the associated robustness and accuracy. Currently, models are optimal in finding relationships between different data modalities to link the patterns that can predict AD diagnosis and progression to identify different types of the disease. Future inclination is preferred in relying on non-invasive screening tests innovations.

Link to the paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8391160/