A mathematical theory of relational generalization in transitive inference.

Humans and animals routinely infer relations between different items or events and generalize these relations to novel combinations of items. This allows them to respond appropriately to radically novel circumstances and is fundamental to advanced cognition. However, how learning systems (including the brain) can implement the necessary inductive biases has been unclear. We investigated transitive inference (TI), a classic relational task paradigm in which subjects must learn a relation ([Formula: see text] and [Formula: see text]) and generalize it to new combinations of items ([Formula: see text]). Through mathematical analysis, we found that a broad range of biologically relevant learning models (e.g. gradient flow or ridge regression) perform TI successfully and recapitulate signature behavioral patterns long observed in living subjects. First, we found that models with item-wise additive representations automatically encode transitive relations. Second, for more general representations, a single scalar "conjunctivity factor" determines model behavior on TI and, further, the principle of norm minimization (a standard statistical inductive bias) enables models with fixed, partly conjunctive representations to generalize transitively. Finally, neural networks in the "rich regime," which enables representation learning and improves generalization on many tasks, unexpectedly show poor generalization and anomalous behavior on TI. We find that such networks implement a form of norm minimization (over hidden weights) that yields a local encoding mechanism lacking transitivity. Our findings show how minimal statistical learning principles give rise to a classical relational inductive bias (transitivity), explain empirically observed behaviors, and establish a formal approach to understanding the neural basis of relational abstraction.

CRISPR/Cas9-generated mutations in a sugar transporter gene reduce cassava susceptibility to bacterial blight.

Bacteria from the genus Xanthomonas are prolific phytopathogens that elicit disease in over 400 plant species. Xanthomonads carry a repertoire of specialized proteins called transcription activator-like (TAL) effectors that promote disease and pathogen virulence by inducing expression of host susceptibility (S) genes. Xanthomonas phaseoli pv. manihotis (Xpm) causes bacterial blight on the staple food crop cassava (Manihot esculenta Crantz). The Xpm effector TAL20 induces ectopic expression of the S gene Manihot esculenta Sugars Will Eventually be Exported Transporter 10a (MeSWEET10a), which encodes a sugar transporter that contributes to cassava bacterial blight susceptibility. We used CRISPR/Cas9 to generate multiple cassava lines with edits to the MeSWEET10a TAL20 effector binding site and/or coding sequence. In several of the regenerated lines, MeSWEET10a expression was no longer induced by Xpm, and in these cases, we observed reduced cassava bacterial blight (CBB) disease symptoms post Xpm infection. Because MeSWEET10a is expressed in cassava flowers, we further characterized the reproductive capability of the MeSWEET10a promoter and coding sequence mutants. Lines were crossed to themselves and to wild-type plants. The results indicated that expression of MeSWEET10a in female, but not male, flowers, is critical to produce viable F1 seed. In the case of promoter mutations that left the coding sequence intact, viable F1 progeny were recovered. Taken together, these results demonstrate that blocking MeSWEET10a induction is a viable strategy for decreasing cassava susceptibility to CBB and that ideal lines will contain promoter mutations that block TAL effector binding while leaving endogenous expression of MeSWEET10a unaltered.

Emergent neural dynamics and geometry for generalization in a transitive inference task.

Relational cognition-the ability to infer relationships that generalize to novel combinations of objects-is fundamental to human and animal intelligence. Despite this importance, it remains unclear how relational cognition is implemented in the brain due in part to a lack of hypotheses and predictions at the levels of collective neural activity and behavior. Here we discovered, analyzed, and experimentally tested neural networks (NNs) that perform transitive inference (TI), a classic relational task (if A > B and B > C, then A > C). We found NNs that (i) generalized perfectly, despite lacking overt transitive structure prior to training, (ii) generalized when the task required working memory (WM), a capacity thought to be essential to inference in the brain, (iii) emergently expressed behaviors long observed in living subjects, in addition to a novel order-dependent behavior, and (iv) expressed different task solutions yielding alternative behavioral and neural predictions. Further, in a large-scale experiment, we found that human subjects performing WM-based TI showed behavior inconsistent with a class of NNs that characteristically expressed an intuitive task solution. These findings provide neural insights into a classical relational ability, with wider implications for how the brain realizes relational cognition.

Superstitious learning of abstract order from random reinforcement.

Humans and other animals often infer spurious associations among unrelated events. However, such superstitious learning is usually accounted for by conditioned associations, raising the question of whether an animal could develop more complex cognitive structures independent of reinforcement. Here, we tasked monkeys with discovering the serial order of two pictorial sets: a "learnable" set in which the stimuli were implicitly ordered and monkeys were rewarded for choosing the higher-rank stimulus and an "unlearnable" set in which stimuli were unordered and feedback was random regardless of the choice. We replicated prior results that monkeys reliably learned the implicit order of the learnable set. Surprisingly, the monkeys behaved as though some ordering also existed in the unlearnable set, showing consistent choice preference that transferred to novel untrained pairs in this set, even under a preference-discouraging reward schedule that gave rewards more frequently to the stimulus that was selected less often. In simulations, a model-free reinforcement learning algorithm (Q-learning) displayed a degree of consistent ordering among the unlearnable set but, unlike the monkeys, failed to do so under the preference-discouraging reward schedule. Our results suggest that monkeys infer abstract structures from objectively random events using heuristics that extend beyond stimulus-outcome conditional learning to more cognitive model-based learning mechanisms.

Influence of Rule- and Reward-based Strategies on Inferences of Serial Order by Monkeys.

Knowledge of transitive relationships between items can contribute to learning the order of a set of stimuli from pairwise comparisons. However, cognitive mechanisms of transitive inferences based on rank order remain unclear, as are relative contributions of reward associations and rule-based inference. To explore these issues, we created a conflict between rule- and reward-based learning during a serial ordering task. Rhesus macaques learned two lists, each containing five stimuli that were trained exclusively with adjacent pairs. Selection of the higher-ranked item resulted in rewards. "Small reward" lists yielded two drops of fluid reward, whereas "large reward" lists yielded five drops. Following training of adjacent pairs, monkeys were tested on novels pairs. One item was selected from each list, such that a ranking rule could conflict with preferences for large rewards. Differences between the corresponding reward magnitudes had a strong influence on accuracy, but we also observed a symbolic distance effect. That provided evidence of a rule-based influence on decisions. RT comparisons suggested a conflict between rule- and reward-based processes. We conclude that performance reflects the contributions of two strategies and that a model-based strategy is employed in the face of a strong countervailing reward incentive.

Positional inference in rhesus macaques.

Understanding how organisms make transitive inferences is critical to understanding their general ability to learn serial relationships. In this context, transitive inference (TI) can be understood as a specific heuristic that applies broadly to many different serial learning tasks, which have been the focus of hundreds of studies involving dozens of species. In the present study, monkeys learned the order of 7-item lists of photographic stimuli by trial and error, and were then tested on "derived" lists. These derived test lists combined stimuli from multiple training lists in ambiguous ways, sometimes changing their order relative to training. We found that subjects displayed strong preferences when presented with novel test pairs, even when those pairs were drawn from different training lists. These preferences were helpful when test pairs had an ordering congruent with their ranks during training, but yielded consistently below-chance performance when pairs had an incongruent order relative to training. This behavior can be explained by the joint contributions of transitive inference and another heuristic that we refer to as "positional inference." Positional inferences play a complementary role to transitive inferences in facilitating choices between novel pairs of stimuli. The theoretical framework that best explains both transitive and positional inferences is a spatial model that represents both the position of each stimulus and its uncertainty. A computational implementation of this framework yields accurate predictions about both correct responses and errors on derived lists.

Category learning in a transitive inference paradigm.

The implied order of a ranked set of visual images can be learned without reliance on information that explicitly signals their order. Such learning is difficult to explain by associative mechanisms, but can be accounted for by cognitive representations and processes such as transitive inference. Our study sought to determine if those processes also apply to learning categories of images. We asked whether participants can (a) infer that stimulus images belonged to familiar categories, even when the images for each trial were unique, and (b) sort those categories into an ordering that obeys transitivity. Participants received minimal verbal instruction and a single session of training. Despite this, they learned the implied order of lists of fixed stimuli and lists of ordered categories, using trial-unique exemplars. We trained two groups, one for which stimuli were constant throughout training and testing (n = 60), and one for which exemplars of each category were trial-unique (n = 50). Our findings suggest that differing cognitive processes may underpin serial learning when learning about specific stimuli as opposed to stimulus categories.

Integrated genomic characterization of ERBB2/HER2 alterations in invasive breast carcinoma: a focus on unusual FISH groups.

In patients with invasive breast cancer, fluorescence in situ hybridization (FISH) testing for HER2 typically demonstrates the clear presence or lack of ERBB2 (HER2) amplification (i.e., groups 1 or 5). However, a small subset of patients can present with unusual HER2 FISH patterns (groups 2-4), resulting in diagnostic confusion. To provide clarity, the 2018 CAP/ASCO HER2 testing guideline recommends additional testing using HER2 immunohistochemistry (IHC) for determining the final HER2 status. Despite this effort, the genomic correlates of unusual HER2 FISH groups remain poorly understood. Here, we used droplet digital PCR (ddPCR) and targeted next-generation sequencing (NGS) to characterize the genomic features of both usual and unusual HER2 FISH groups. In this study, 51 clinical samples were selected to represent FISH groups 1-5. Furthermore, group 1 was subdivided into two groups, with groups 1A and 1B corresponding to cases with HER2 signals/cell ≥6.0 and 4-6, respectively. Overall, our findings revealed a wide range of copy number alterations in HER2 across the different FISH groups. As expected, groups 1A and 5 showed the clear presence and lack of HER2 copy number gain, respectively, as measured by ddPCR and NGS. In contrast, group 1B and other uncommon FISH groups (groups 2-4) were characterized by a broader range of HER2 copy levels with only a few select cases showing high-level gain. Notably, these cases with increased HER2 copy levels also showed HER2 overexpression by IHC, thus highlighting the correlation between HER2 copy number and HER2 protein expression. Given the concordance between the genomic and protein results, our findings suggest that HER2 IHC may inform HER2 copy number status in patients with unusual FISH patterns. Hence, our results support the current recommendation for using IHC to resolve HER2 status in FISH groups 2-4.

Use of gene knockout to examine serotonergic control of ion uptake in zebrafish reveals the importance of controlling for genetic background: A cautionary tale.

Freshwater (FW) fishes inhabit dilute environments and must actively absorb ions in order to counteract diffusive salt loss. Neuroendocrine control of ion uptake in FW fishes is an important feature of ion homeostasis and several important neuroendocrine factors have been identified. The role of serotonin (5-HT), however, has received less attention despite several studies pointing to a role for 5-HT in the control of ion balance. Here, we used a gene knockout approach to elucidate the role of 5-HT in regulating Na+ and Ca2+ uptake rates in larval zebrafish. Tryptophan hydroxylase (TPH) is the rate-limiting step in 5-HT synthesis and we therefore hypothesized that ion uptake rates would be altered in zebrafish larvae carrying knockout mutations in tph genes. We first examined the effect of tph1b knockout (KO) and found that tph1bKO larvae, obtained from Harvard University, had reduced rates of Na+ and Ca2+ uptake compared to wild-type (WT) larvae from our institution (uOttawa WT), lending support to our hypothesis. However, further experiments controlling for differences in genetic background demonstrated that WT larvae from Harvard University (Harvard WT) had lower ion uptake rates than those of uOttawa WT, and that ion uptake rate between Harvard WT and tph1bKO larvae were not significantly different. Therefore, our initial observation that tph1bKO larvae (Harvard source) had reduced ion uptake rates relative to uOttawa WT was a function of genetic background and not of knockout itself. These data provide a cautionary tale of the importance of controlling for genetic background in gene knockout experiments.

Inferential Learning of Serial Order of Perceptual Categories by Rhesus Monkeys (Macaca mulatta).

Category learning in animals is typically trained explicitly, in most instances by varying the exemplars of a single category in a matching-to-sample task. Here, we show that male rhesus macaques can learn categories by a transitive inference paradigm in which novel exemplars of five categories were presented throughout training. Instead of requiring decisions about a constant set of repetitively presented stimuli, we studied the macaque's ability to determine the relative order of multiple exemplars of particular stimuli that were rarely repeated. Ordinal decisions generalized both to novel stimuli and, as a consequence, to novel pairings. Thus, we showed that rhesus monkeys could learn to categorize on the basis of implied ordinal position, without prior matching-to-sample training, and that they could then make inferences about category order. Our results challenge the plausibility of association models of category learning and broaden the scope of the transitive inference paradigm.SIGNIFICANCE STATEMENT The cognitive abilities of nonhuman animals are of enduring interest to scientists and the general public because they blur the dividing line between human and nonhuman intelligence. Categorization and sequence learning are highly abstract cognitive abilities each in their own right. This study is the first to provide evidence that visual categories can be ordered serially by macaque monkeys using a behavioral paradigm that provides no explicit feedback about category or serial order. These results strongly challenge accounts of learning based on stimulus-response associations.

Implicit Value Updating Explains Transitive Inference Performance: The Betasort Model.

Transitive inference (the ability to infer that B > D given that B > C and C > D) is a widespread characteristic of serial learning, observed in dozens of species. Despite these robust behavioral effects, reinforcement learning models reliant on reward prediction error or associative strength routinely fail to perform these inferences. We propose an algorithm called betasort, inspired by cognitive processes, which performs transitive inference at low computational cost. This is accomplished by (1) representing stimulus positions along a unit span using beta distributions, (2) treating positive and negative feedback asymmetrically, and (3) updating the position of every stimulus during every trial, whether that stimulus was visible or not. Performance was compared for rhesus macaques, humans, and the betasort algorithm, as well as Q-learning, an established reward-prediction error (RPE) model. Of these, only Q-learning failed to respond above chance during critical test trials. Betasort's success (when compared to RPE models) and its computational efficiency (when compared to full Markov decision process implementations) suggests that the study of reinforcement learning in organisms will be best served by a feature-driven approach to comparing formal models.

Acquisition of conditioned responding in a multiple schedule depends on the reinforcement's temporal contingency with each stimulus.

Forty mice acquired conditioned responses to stimuli presented in a multiple schedule with variable inter-trial intervals (ITIs). In some trials, reinforcement was preceded by a variable conditioned stimulus (CS), while other trials were reinforced following distinctive fixed-duration CS. A third stimulus was presented but never paired with reinforcement. Subjects in five groups experienced ITIs of different durations. Acquisition of responding to each stimulus depended only on the cycle-to-trial ratio (C/T), and thus on the temporal contingency of each stimulus. Acquisition was unaffected by whether CSs were of fixed or variable duration.

Rapid cognitive flexibility of rhesus macaques performing psychophysical task-switching.

Three rhesus monkeys (Macaca mulatta) performed a simultaneous chaining task in which stimuli had to be sorted according to their visual properties. Each stimulus could vary independently along two dimensions (luminosity and radius), and a cue indicating which dimension to sort by was random trial to trial. These rapid and unpredictable changes constitute a task-switching paradigm, in which subjects must encode task demands and shift to whichever task-set is presently activated. In contrast to the widely reported task-switching delay observed in human studies, our subjects show no appreciable reduction in reaction times following a switch in the task requirements. Also, in contrast to the results of studies on human subjects, monkeys experienced enduring interference from trial-irrelevant stimulus features, even after exhaustive training. These results are consistent with a small but growing body of evidence that task-switching in rhesus macaques differs in basic ways from the pattern of behavior reported in studies of human cognition. Given the importance of task-switching paradigms in cognitive and clinical assessment, and the frequency with which corresponding animal models rely on non-human primates, understanding these differences in behavior is essential to the comparative study of cognitive impairment.

Gamification and motivation: Impact on delay discounting performance.

Delay discounting is a phenomenon strongly associated with impulsivity. However, in order for a measured discounting rate in an experiment to meaningfully generalize to choices made elsewhere in life, participants must provide thoughtful, engaged answers during the assessment. Classic discounting tasks may not optimize intrinsic motivation or enjoyment, and a participant who is disengaged from the task is likely to behave in a way that provides a biased estimate of their discounting function. We assessed degree of delay discounting in a task intended to vary level of participant motivation. This was accomplished by introducing varying levels of gamification, the application of game design principles to a non-game context. Experiment 1 compared three versions of the delay discounting task with differing degrees of gamification and compared performance and task enjoyment across those variations, while Experiment 2 used two conditions (one gamified, one not). Participants found more gamified versions of the task more enjoyable than the other conditions, without producing substantial between-group differences in most cases. Thus, more polished task gameplay can provide a more enjoyable experience for participants without undermining delay discounting effects commonly reported in the literature. We also found that in all experimental conditions, higher levels of interest in or enjoyment of the task tended to be associated with more rapid discounting. This may suggest that low task motivation may result in less impulsive choice and suggests that participants who find delay discounting experiments sufficiently boring may bias assessments of value across delays.

A mathematical theory of relational generalization in transitive inference.

Humans and animals routinely infer relations between different items or events and generalize these relations to novel combinations of items. This allows them to respond appropriately to radically novel circumstances and is fundamental to advanced cognition. However, how learning systems (including the brain) can implement the necessary inductive biases has been unclear. Here we investigated transitive inference (TI), a classic relational task paradigm in which subjects must learn a relation (A > B and B > C) and generalize it to new combinations of items (A > C). Through mathematical analysis, we found that a broad range of biologically relevant learning models (e.g. gradient flow or ridge regression) perform TI successfully and recapitulate signature behavioral patterns long observed in living subjects. First, we found that models with item-wise additive representations automatically encode transitive relations. Second, for more general representations, a single scalar "conjunctivity factor" determines model behavior on TI and, further, the principle of norm minimization (a standard statistical inductive bias) enables models with fixed, partly conjunctive representations to generalize transitively. Finally, neural networks in the "rich regime," which enables representation learning and has been found to improve generalization, unexpectedly show poor generalization and anomalous behavior. We find that such networks implement a form of norm minimization (over hidden weights) that yields a local encoding mechanism lacking transitivity. Our findings show how minimal statistical learning principles give rise to a classical relational inductive bias (transitivity), explain empirically observed behaviors, and establish a formal approach to understanding the neural basis of relational abstraction.

Monkeys would rather see and do: preference for agentic control in rhesus macaques.

Rhesus monkeys (Macaca mulatta) engaged in a series of computerized tasks modeled on billiards and arcade games in order to determine their degree of preference for scenarios in which food rewards were contingent on their actions, as opposed to those in which outcomes appeared externally caused. Throughout these tasks, subjects showed a consistent preference for "agentic control," a state in which goal-directed behavior is directly responsible for motivating outcomes. Other factors like the frequency and timing of reward deliveries were precisely controlled and did not explain observed preferences.

Inefficient prioritization of task-relevant attributes during instrumental information demand.

In natural settings, people evaluate complex multi-attribute situations and decide which attribute to request information about. Little is known about how people make this selection and specifically, how they identify individual observations that best predict the value of a multi-attribute situation. Here show that, in a simple task of information demand, participants inefficiently query attributes that have high individual value but are relatively uninformative about a total payoff. This inefficiency is robust in two instrumental conditions in which gathering less informative observations leads to significantly lower rewards. Across individuals, variations in the sensitivity to informativeness is associated with personality metrics, showing negative associations with extraversion and thrill seeking and positive associations with stress tolerance and need for cognition. Thus, people select informative queries using sub-optimal strategies that are associated with personality traits and influence consequential choices.

Improving cassava bacterial blight resistance by editing the epigenome.

Pathogens rely on expression of host susceptibility (S) genes to promote infection and disease. As DNA methylation is an epigenetic modification that affects gene expression, blocking access to S genes through targeted methylation could increase disease resistance. Xanthomonas phaseoli pv. manihotis, the causal agent of cassava bacterial blight (CBB), uses transcription activator-like20 (TAL20) to induce expression of the S gene MeSWEET10a. In this work, we direct methylation to the TAL20 effector binding element within the MeSWEET10a promoter using a synthetic zinc-finger DNA binding domain fused to a component of the RNA-directed DNA methylation pathway. We demonstrate that this methylation prevents TAL20 binding, blocks transcriptional activation of MeSWEET10a in vivo and that these plants display decreased CBB symptoms while maintaining normal growth and development. This work therefore presents an epigenome editing approach useful for crop improvement.

Disruption of tph1 genes demonstrates the importance of serotonin in regulating ventilation in larval zebrafish (Danio rerio).

Serotonergic neuroepithelial cells (NECs) in larval zebrafish are believed to be O2 chemoreceptors. Serotonin (5-HT) within these NECs has been implicated as a neurotransmitter mediating the hypoxic ventilatory response (HVR). Here, we use knockout approaches to discern the role of 5-HT in regulating the HVR by targeting the rate limiting enzyme for 5-HT synthesis, tryptophan hydroxylase (Tph). Using transgenic lines, we determined that Tph1a is expressed in skin and pharyngeal arch NECs, as well as in pharyngeal arch Merkel-like cells (MLCs), whereas Tph1b is expressed predominately in MLCs. Knocking out the two tph1 paralogs resulted in similar changes in detectable serotonergic cell density between the two mutants, yet their responses to hypoxia (35 mmHg) were different. Larvae lacking Tph1a (tph1a-/- mutants) displayed a higher ventilation rate when exposed to hypoxia compared to wild-types, whereas tph1b-/- mutants exhibited a lower ventilation rate suggesting that 5-HT located in locations other than NECs, may play a dominant role in regulating the HVR.

Failure to Find Altruistic Food Sharing in Rats.

Prior research has found that one rat will release a second rat from a restraint in the presence of food, thereby allowing that second rat access to food. Such behavior, clearly beneficial to the second rat and costly to the first, has been interpreted as altruistic. Because clear demonstrations of altruism in rats are rare, such findings deserve a careful look. The present study aimed to replicate this finding, but with more systematic methods to examine whether, and under what conditions, a rat might share food with its cagemate partner. Rats were given repeated choices between high-valued food (sucrose pellets) and 30-s social access to a familiar rat, with the (a) food size (number of food pellets per response), and (b) food motivation (extra-session access to food) varied across conditions. Rats responded consistently for both food and social interaction, but at different levels and with different sensitivity to the food-access manipulations. Food production and consumption was high when food motivation was also high (food restriction) but substantially lower when food motivation was low (unlimited food access). Social release occurred at moderate levels, unaffected by the food-based manipulations. When food was abundant and food motivation low, the rats chose food and social options about equally often, but sharing (food left unconsumed prior to social release) occurred at low levels across sessions and conditions. Even under conditions of low food motivation, sharing occurred on only 1% of the sharing opportunities. The results are therefore inconsistent with claims in the literature that rats are altruistically motivated to share food with other rats.