Andrew Whalen

The coevolution of innovation and technical intelligence in primates

In birds and primates, the frequency of behavioural innovation has been shown to covary with absolute and relative brain size, leading to the suggestion that large brains allow animals to innovate, and/or that selection for innovativeness, together with social learning, may have driven brain enlargement. We examined the relationship between primate brain size and both technical (i.e. tool using) and non-technical innovation, deploying a combination of phylogenetically informed regression and exploratory causal graph analyses. Regression analyses revealed that absolute and relative brain size correlated positively with technical innovation, and exhibited consistently weaker, but still positive, relationships with non-technical innovation. These findings mirror similar results in birds. Our exploratory causal graph analyses suggested that technical innovation shares strong direct relationships with brain size, body size, social learning rate and social group size, whereas non-technical innovation did not exhibit a direct relationship with brain size. Nonetheless, non-technical innovation was linked to brain size indirectly via diet and life-history variables. Our findings support ‘technical intelligence’ hypotheses in linking technical innovation to encephalization in the restricted set of primate lineages where technical innovation has been reported. Our findings also provide support for a broad co-evolving complex of brain, behaviour, life-history, social and dietary variables, providing secondary support for social and ecological intelligence hypotheses. The ability to gain access to difficult-to-extract, but potentially nutrient-rich, resources through tool use may have conferred on some primates adaptive advantages, leading to selection for brain circuitry that underlies technical proficiency.

The learning of action sequences through social transmission

Previous empirical work on animal social learning has found that many species lack the ability to learn entire action sequences solely through reliance on social information. Conversely, acquiring action sequences through asocial learning can be difficult due to the large number of potential sequences arising from even a small number of base actions. In spite of this, several studies report that some primates use action sequences in the wild. We investigate how social information can be integrated with asocial learning to facilitate the learning of action sequences. We formalize this problem by examining how learners using temporal difference learning, a widely applicable model of reinforcement learning, can combine social cues with their own experiences to acquire action sequences. The learning problem is modeled as a Markov decision process. The learning of nettle processing by mountain gorillas serves as a focal example. Through simulations, we find that the social facilitation of component actions can combine with individual learning to facilitate the acquisition of action sequences. Our analysis illustrates that how even simple forms of social learning, combined with asocial learning, generate substantially faster learning of action sequences compared to asocial processes alone, and that the benefits of social information increase with the length of the action sequence and the number of base actions.

Conformity biased transmission in social networks

In this paper we explore how the structure of a population can differentially influence the spread of novel behaviors, depending on the learning strategy of each individual. We use a series of simulations to analyze how frequency dependent learning rules might affect how easily novel behaviors can spread through a population on four artificial social networks, and three real social networks. We measured the likelihood that a novel behavior could spread through the population, and the likelihood that there were multiple behavioral variants in the population, a measure of cultural diversity. Surprisingly, we find few differences between networks on either measure. However, we do find that where a behavior originated on a network can have a substantial impact on the likelihood that it spreads, and that this location effect depends on the learning strategy of an individual. These results suggest that for first-order analysis of how behaviors spread through a population, social network structure can be ignored, but that the social network structure may be useful for more fine-tuned analyses and predictions.

Fish pool their experience to solve problems collectively

Access to information is a key advantage of grouping. Although experienced animals can lead others to solve problems, less is known about whether partially informed individuals can pool experiences to overcome challenges collectively. Here we provide evidence of such ‘experience-pooling’. We presented shoals of sticklebacks (Gasterosteus aculeatus) with a two-stage foraging task requiring them to find and access hidden food. Individual fish were either inexperienced or had knowledge of just one of the stages. Shoals containing individuals trained in each of the stages pooled their expertise, allowing more fish to access the food, and to do so more rapidly, compared with other shoal compositions. Strong social effects were identified: the presence of experienced individuals increased the likelihood of untrained fish completing each stage. These findings demonstrate that animal groups can integrate individual experience to solve multi-stage problems, and have implications for our understanding of social foraging, migration and social systems.

Bayesian Model Selection with Network Based Diffusion Analysis

A number of recent studies have used Network Based Diffusion Analysis (NBDA) to detect the role of social transmission in the spread of a novel behavior through a population. In this paper we present a unified framework for performing NBDA in a Bayesian setting, and demonstrate how the Watanabe Akaike Information Criteria (WAIC) can be used for model selection. We present a specific example of applying this method to Time to Acquisition Diffusion Analysis (TADA). To examine the robustness of this technique, we performed a large scale simulation study and found that NBDA using WAIC could recover the correct model of social transmission under a wide range of cases, including under the presence of random effects, individual level variables, and alternative models of social transmission. This work suggests that NBDA is an effective and widely applicable tool for uncovering whether social transmission underpins the spread of a novel behavior, and may still provide accurate results even when key model assumptions are relaxed.

Assessment of the performance of different hidden Markov models for imputation in animal breeding

In this paper we review the performance of various hidden Markov model-based imputation methods in animal breeding populations. Traditionally, heuristic-based imputation methods have been used for imputation in large animal populations due to their computational efficiency, scalability, and accuracy. However, recent advances in the area of human genetics have increased the ability of probabilistic hidden Markov model methods to perform accurate phasing and imputation in large populations. These advances may enable these methods to be useful for routine use in large animal populations. To test this, we evaluate here the accuracy and computational cost of several methods in a series of simulated populations and a real animal population. We first tested single-step (diploid) imputation, which performs both phasing and imputation. Then we tested pre-phasing followed by haploid imputation. We tested four diploid imputation methods (fastPHASE, Beagle v4.0, IMPUTE2, and MaCH), three phasing methods, (SHAPEIT2, HAPI-UR, and Eagle2), and three haploid imputation methods (IMPUTE2, Beagle v4.1, and minimac3). We found that performing pre-phasing and haploid imputation was faster and more accurate than diploid imputation. In particular, we found that pre-phasing with Eagle2 or HAPI-UR and imputing with minimac3 or IMPUTE2 gave the highest accuracies in both simulated and real data.

Parentage assignment with low density array data and low coverage sequence data

In this paper we developed and analysed a parentage assignment method for low density array data or low coverage sequence data. We tested the algorithm on a series of simulated datasets varying the genotyping error rate, number of potential sires, and availability of maternal information. We found that for array data, to obtain more than 99% sire assignment accuracy 250 SNPs were sufficient when the dam was unknown and 100 SNPs were sufficient when the dam was known and genotyped. These results replicated on a real pig dataset. We found that for sequence data, to obtain 99% sire assignment accuracy 500 reads on genotyped sites were sufficient when the dam was unknown, and 200 reads were sufficient when the dam was known. We also investigated the benefit of pre-correcting genotyping errors using an imputation algorithm, and utilizing linkage information to increase accuracy. We found that pre-correcting genotyping errors could drastically increase parentage assignment accuracy, but that there was little benefit to using linkage information when typed or sequence loci were distributed across multiple chromosomes. We provide this parentage assignment method as a standalone program called AlphaAssign.In this paper we evaluate using genotype-by-sequencing (GBS) data to perform parentage assignment in lieu of traditional array data. The use of GBS data raises two issues: First, for low-coverage GBS data, it may not be possible to call the genotype at many loci, a critical first step for detecting opposing homozygous markers. Second, the amount of sequencing coverage may vary across individuals, making it challenging to directly compare the likelihood scores between putative parents. To address these issues we extend the probabilistic framework of Huisman (2017) and evaluate putative parents by comparing their (potentially noisy) genotypes to a series of proposal distributions. These distributions describe the expected genotype probabilities for the relatives of an individual. We assign putative parents as a parent if they are classified as a parent (as opposed to e.g., an unrelated individual), and if the assignment score passes a threshold. We evaluated this method on simulated data and found that (1) high-coverage GBS data performs similarly to array data and requires only a small number of markers to correctly assign parents and (2) low-coverage GBS data (as low as 0.1x) can also be used, provided that it is obtained across a large number of markers. When analysing the low-coverage GBS data, we also found a high number of false positives if the true parent is not contained within the list of candidate parents, but that this false positive rate can be greatly reduced by hand tuning the assignment threshold. We provide this parentage assignment method as a standalone program called AlphaAssign.

Social learning strategies regulate the wisdom and madness of interactive online crowds

Decentralised social interactions can generate swarm intelligence, but may concurrently increase the risk of maladaptive herding. Here we present an individual-based model analysis suggesting that the conflict between the ‘wisdom’ and ‘madness’ of interactive crowds is regulated by selectively choosing which social learning strategy to use. We used an interactive online experiment with 699 participants to measure the patterns of human social-information use, varying both task uncertainty and group size. Hierarchical Bayesian analyses identified the individual learning strategies, revealing that conformity bias increased with the task’s uncertainty, whereas reliance on social learning increased with group size. Mapping individual strategies onto collective behaviour, we show that maladaptive herding occurred more frequently when larger groups were engaged in more uncertain tasks. Our computational modelling approach provides novel evidence that the likelihood of swarm intelligence versus herding can be predicted using knowledge of social learning strategies.

Assessment of the performance of hidden Markov models for imputation in animal breeding

BackgroundIn this paper, we review the performance of various hidden Markov model-based imputation methods in animal breeding populations. Traditionally, pedigree and heuristic-based imputation methods have been used for imputation in large animal populations due to their computational efficiency, scalability, and accuracy. Recent advances in the area of human genetics have increased the ability of probabilistic hidden Markov model methods to perform accurate phasing and imputation in large populations. These advances may enable these methods to be useful for routine use in large animal populations, particularly in populations where pedigree information is not readily available.MethodsTo test the performance of hidden Markov model-based imputation, we evaluated the accuracy and computational cost of several methods in a series of simulated populations and a real animal population without using a pedigree. First, we tested single-step (diploid) imputation, which performs both phasing and imputation. Second, we tested pre-phasing followed by haploid imputation. Overall, we used four available diploid imputation methods (fastPHASE, Beagle v4.0, IMPUTE2, and MaCH), three phasing methods, (SHAPEIT2, HAPI-UR, and Eagle2), and three haploid imputation methods (IMPUTE2, Beagle v4.1, and Minimac3).ResultsWe found that performing pre-phasing and haploid imputation was faster and more accurate than diploid imputation. In particular, among all the methods tested, pre-phasing with Eagle2 or HAPI-UR and imputing with Minimac3 or IMPUTE2 gave the highest accuracies with both simulated and real data.ConclusionsThe results of this study suggest that hidden Markov model-based imputation algorithms are an accurate and computationally feasible approach for performing imputation without a pedigree when pre-phasing and haploid imputation are used. Of the algorithms tested, the combination of Eagle2 and Minimac3 gave the highest accuracy across the simulated and real datasets.

Hybrid peeling for fast and accurate calling, phasing, and imputation with sequence data of any coverage in pedigrees

In this paper we extend multi-locus iterative peeling to be a computationally efficient method for calling, phasing, and imputing sequence data of any coverage in small or large pedigrees. Our method, called hybrid peeling, uses multi-locus iterative peeling to estimate shared chromosome segments between parents and their offspring, and then uses single-locus iterative peeling to aggregate genomic information across multiple generations. Using a synthetic dataset, we first analysed the performance of hybrid peeling for calling and phasing alleles in disconnected families, families which contained only a focal individual and its parents and grandparents. Second, we analysed the performance of hybrid peeling for calling and phasing alleles in the context of the full pedigree. Third, we analysed the performance of hybrid peeling for imputing whole genome sequence data to the remaining individuals in the population. We found that hybrid peeling substantially increase the number of genotypes that were called and phased by leveraging sequence information on related individuals. The calling rate and accuracy increased when the full pedigree was used compared to a reduced pedigree of just parents and grandparents. Finally, hybrid peeling accurately imputed whole genome sequence information to non-sequenced individuals. We believe that this algorithm will enable the generation of low cost and high accuracy whole genome sequence data in many pedigreed populations. We are making this algorithm available as a standalone program called AlphaPeel.