Modern genetics has developed a powerful method called genomic selection that uses the data contained in the genomes of plants and animals to improve breeding.
Genomic selection enables the prediction of an individual’s genetic potential for desired qualities, such as disease resistance, yield, or quality, by examining the DNA sequence differences across individuals.
This genetic data helps to make better-informed decisions, speed up the selection process, and create breeding programs that are more effective and fruitful.
In this interesting area of genetics, machine learning has been creating wonders and enhancing the field. Let’s delve into and learn about machine learning in genomic selection.
What Exactly is Genomic Selection?
Genomic selection is a technique used in animal and plant breeding to forecast an individual’s performance based on their genetic makeup.
It involves looking at people’s DNA to find particular markers connected to desirable characteristics.
Researchers can determine a person’s genetic potential for features like disease resistance, yield, or quality by analyzing these markers across the entire genome.
Breeders can forecast the performance of offspring more accurately thanks to genomic selection without the need for time-consuming and expensive phenotypic assessments.
By enabling breeders to choose individuals with the best genetic potential for breeding programs, this method aids in the acceleration of the breeding process by enabling more effective and focused enhancement of desired traits in plant and animal populations.
Plant breeding via genomic selection
Plant breeding has undergone a revolution thanks to genomic selection, which has sped up the process and increased crop yields.
But to address the oncoming problems brought on by climate change, more development is necessary.
To solve this, researchers are utilizing pangenomes and cutting-edge machine-learning approaches in genomic selection.
The whole genomic material of a species, also known as the pangenome, allows for a thorough comprehension of genetic variation.
We can open the way for crop improvement and mitigate the detrimental effects of climate change on agriculture by looking at examples from crop breeding, comprehending the constraints of machine learning, and highlighting the promise of these techniques.
Pangenomes of Plants: Revealing Genomic Diversity
Traditionally, single-reference genome assemblies have been the primary focus of genomic selection, but pangenomes are now becoming more prevalent. Plant pangenomes, rather than individual genome assemblies, reflect the genetic material of a species or family.
Significant gene variants, including those not included in the reference assembly, are revealed by them. For several crops, pangenomes have been created, illuminating the history of plant domestication and breeding.
Their combination with genetic selection is yet only partially effective.
Breeders can use a wider variety of genetic markers, improving prediction accuracy and capturing all potential connections, by combining pangenomes into genomic selection.
Genomic Selection Based on Machine Learning
Traditional genomic selection approaches have difficulties in addressing nonadditive effects like epistasis, genomic imprinting, and genotype interactions. By simulating these impacts, machine learning approaches provide viable answers.
Recent studies have used machine learning methods in genomic selection, with results ranging between datasets and crops.
Machine learning algorithms are capable of handling complicated data representations, such as mixed phenotypes and interactions between phenotypes or genotypes.
For example, machine learning algorithms have been used to predict production and fruit quality features in polyploid crops such as strawberries and blueberries.
While these systems have great potential, understanding their interpretability and adjusting hyperparameters are critical for effective application.
Different Methods of Machine Learning
In genomic prediction studies, the use of machine learning techniques is growing. These techniques can be separated into ways for supervised and unsupervised learning.
Methods of supervised learning are particularly useful since they can discover patterns in labeled data and anticipate outcomes.
While various studies have examined the prediction effectiveness of specific machine learning approaches, research comparing diverse sets of methods is lacking.
It is critical to understand which groupings of methods perform better and to weigh their benefits and drawbacks in comparison with conventional ways.
Promising Genomic Prediction Methods
Linear Mixed Models
In genomic prediction, conventional linear mixed models have proven to be trustworthy and useful. To account for genetic variation in the population, these models integrate both fixed and random effects.
These algorithms can accurately predict genomic breeding values by taking individual relatedness into account.
Because of their competitive predictive performance, computational efficiency, and simplicity, linear mixed models are widely utilized in plant and animal breeding. They require fewer tuning parameters than other approaches, making them suitable for genomic selection.
For genome prediction, regularized regression methods like LASSO (Least Absolute Shrinkage and Selection Operator) and ridge regression are effective tools.
These techniques enable variable selection and regularization by adding a penalty term to the conventional regression model.
These methods efficiently handle high-dimensional data and enhance prediction accuracy by reducing less significant markers toward zero.
Regularized regression techniques are appealing choices for genomic selection in both plant and animal breeding studies because they strike a compromise between simplicity and effectiveness.
An ensemble learning technique called random forests makes predictions using decision trees. Random forests can be used to assess high-dimensional genomic data in the context of genomic prediction.
With this method, a large number of decision trees are built, each trained on a random subset of markers, and their predictions are combined to produce a single forecast.
Random forests are a useful tool for genomic selection because they can identify intricate interactions and nonlinear correlations between characteristics and markers.
Random forests are also resilient to outliers and can accommodate missing data, which increases their value for genomic prediction.
ANNs (artificial neural networks)
Artificial neural networks, sometimes referred to as ANNs or neural networks, are computational models that draw inspiration from the neural architecture of the human brain.
Due to their capacity to recognize intricate patterns and relationships in data, ANNs have become more and more common in genetic prediction.
ANNs can record nonlinear interactions between markers and attributes because of their multilayer architecture and interconnected nodes (neurons). These networks need thorough training using vast datasets and rigorous hyperparameter adjustment.
By revealing complex genetic links and identifying hidden patterns in genomic data, ANNs have the potential to increase the accuracy of genomic prediction.
Target Traits and Importance of Data
Studies show that the particular data and target attributes being evaluated have an impact on the prediction performance and computational costs of machine learning approaches.
As can be observed, adding complexity to traditional regularized approaches can result in large computing costs without necessarily boosting prediction precision.
Computational Efficiency Investments
Given the reliance on target datasets and attributes for predictive performance and computational burden, it is critical to invest in improving the computational efficiency of machine learning algorithms and computing resources.
This would help improve the precision and efficiency of genomic selection.
Conclusion – What Does the Future Hold?
Machine learning in genomic selection seems to have a bright future. Machine learning techniques have the potential to completely change genetic prediction as technology develops and computer resources become more widely available.
These methods allow for the handling of high-dimensional genomic data, the discovery of intricate patterns, and an increase in prediction accuracy.
By facilitating a quicker and more accurate selection of individuals with desired features, the combination of machine learning algorithms with genomic selection holds the possibility of improving breeding programs.
To improve these techniques, deal with computational issues, and investigate their application to different plant and animal species, more study is required.
We expect machine learning to become increasingly important in genomic selection as technology develops, speeding the rate of genetic progress and assisting the agriculsectorture .