Students Deserve Mentors

The best form of education is personal mentoring. More of our educational and work system should get back to that model.

We learned some important things from the Covid pandemic. One is that fiscal stimulus really works. Another is that mRNA vaccines are highly effective, and their rapid and flexible development cycle makes them a superior platform for future vaccines. And another is that social interaction is deeply important, especially for young people. We all got used to Zoom, but for school children, that was a poor substitute, when it was even possible. Children were left significantly behind both academically and socially.

A recent segment on the PBS NewsHour touched on this in a discussion of adolescent development. Its message was that learning requires challenging opportunities and human relationships. Adolescents are going on a heroic quest to become adults. They thrive on active engagement with the world and need models of successful adulthood to learn from. How to provide these key functions in an optimal way? We know how to do this- by apprenticeship and mentorship. This model has been understood forever, from the schools of Athens to the medieval trade guilds to the graduate schools of contemporary academia. My grandfather was a baker in Germany, and in his turn trained many apprentices and journeymen to be bakers. I went to graduate school, which turned out to be a glorified apprenticeship under a renowned researcher, then went on to a journeyman position (aka post-doc) with another mentor. This model is an education in many dimensions- the technical ingredients of a craft, the management practices that make a successful organization, how to participate in a larger community that pursues socially important goals, and the discipline and moral integrity it takes to be a competent adult, capable of leadership.

Example of a certificate of attainment of mastery, 1927, for a bricklayer, attested by his mentors and examiners.

However, as a society, we are reluctant to make these kinds of investments in children and adolescents. Efficiency demands that class sizes be large, colleges impersonal, and money squeezed out of the system. Companies clamor for fully trained job candidates, expecting students to go into debt in trade schools before being hired into a paying job. Few young people get the kind of lengthy, personal training that they would most benefit from. Mentorship becomes a hazard of chance, if a boss in an early job takes an interest, or a teacher decides to make extra time.

Principally, I fault the corporate system, which has sloughed off its civic responsibilities to train people and propagate cultural knowledge. The economy is full of interesting and important jobs representing exquisite technical knowledge and other expertise. As a culture and economy, we are not going to maintain a high standard if we keep losing these skills and knowledge with every generation. Just look what has happened to the industries we have ceded to China. Innovation hubs like Silicon Valley are successful in part because training becomes a shared enterprise. New companies benefit from a large pool of experienced workers, who can switch between organizations with ease. No individual company carries the whole burden of training, but as companies become larger and more specialized, they have to take on the costs of training a larger proportion of their incoming employees. Yet they still benefit from the cross-fertilization of being in a highly skilled employment ecosystem.

To better serve young people, we need to make integration into corporate skills training more accessible and normal. The idea that students should be battling for unpaid internships is absurd and insulting- all internships should be paid, and they should be longer as well. The German trades system is an example, where companies and government cooperate in providing training to young people. The companies get a much better familiarity with future hires, who are also better trained. Many trades/sectors have a communal "training tax", which all companies pay, and which funds salaries to trainees and other training costs. This is one accomplishment of the union system in Germany, which is much stronger and better integrated into their industries than that in the US.

This model could be made more general in the US as a federal program, crossing all organizations in the public and private sector, funding internships and training for more students than is now done, setting up a more lengthy and regular apprenticeship system. The training/salary costs would grade over the first few years of employment from tax-supported to company-supported. Lowering the burden of a young first hire, both in financial terms and terms of knowing the candidates better, should encourage more hiring and more training by employers. 

Companies are often citadels of hermetic wisdom, when they are not going off the rails as predatory enterprises. Integrating more young people and an additional purpose of training into US corporate culture would counteract both of these problems, while helping the youth and preserving / propagating cultural knowledge more effectively.

• Mentoring, Shanghai style.
• Surrender Dorothy!
• Mini-me and the future of Ukraine.
• Free buses work.
• Crypto, on the other hand, remains a criminal enterprise.
• Biology is not so easy. Really.

Ground Truth for Genetic Mutations

Saturation mutagenasis shows that our estimates of the functional effect of uncharacterized mutations are not so great.

Human genomes can now be sequenced for less than $1,000. This technological revolution has enabled a large expansion of genetic testing, used for cancer tissue diagnosis and tracking, and for genetic syndrome analysis both of embryos before birth and affected people after birth. But just because a base among the 3 billion of the genome is different from the "reference" genome, that does not mean it is bad. Judging whether a variant (the modern, more neutral term for mutation) is bad takes a lot of educated guesswork.

A recent paper described a deep dive into one gene, where the authors created and characterized the functional consequence of every possible coding variant. Then they evaluated how well our current rules of thumb and prediction programs for variant analysis compare with what they found. It was a mediocre performance. The gene is CDKN2A, one of our more curious oddities. This is an important tumor suppressor gene that inhibits cell cycle progression and promotes DNA repair- it is often mutated in cancers. But it encodes not one, but two entirely different proteins, by virtue of a complex mRNA splicing pattern that uses distinct exons in some coding portions, and parts of one sequence in two different frames, to encode these two proteins, called p16 and p14. 

One gene, two proteins. CDKN2A has a splicing pattern (mRNA exons shown as boxes at top, with pink segments leading to the p14 product, and the blue segments leading the p16 product) that generates two entirely different proteins from one gene. Each product has tumor suppressing effects, though via distinct mechanisms.

Regardless of the complex splicing and protein coding characteristics, the authors generated all possible variants in every possible coded amino acid (156 amino acids in all, as both produced proteins are relatively short). Since the primary roles of these proteins are in cell cycle and proliferation control, it was possible to assay function by their effect when expressed in cultured pancreatic cells. A deleterious effect on the protein was revealed as, paradoxically, increased growth of these cells. They found that about 600 of the 3,000 different variants in their catalog had such an effect, or 20%.

This is an expected rate of effect, on the whole. Most positions in proteins are not that important, and can be substituted by several similar amino acids. For a typical enzyme, for instance, the active site may be made up of a few amino acids in a particular orientation, and the rest of the protein is there to fold into the required shape to form that active site. Similar folding can be facilitated by numerous amino acids at most positions, as has been richly documented in evolutionary studies of closely-related proteins. These p16 and p14 proteins interact with a few partners, so they need to maintain those key interfacial surfaces to be fully functional. Additionally, the assay these researchers ran, of a few generations of growth, is far less sensitive than a long-term true evolutionary setting, which can sift out very small effects on a protein, so they were setting a relatively high bar for seeing a deleterious effect. They did a selective replication of their own study, and found a reproducibility rate of about 80%, which is not great, frankly.

"Of variants identified in patients with cancer and previously reported to be functionally deleterious in published literature and/or reported in ClinVar as pathogenic or likely pathogenic (benchmark pathogenic variants), 27 of 32 (84.4%) were functionally deleterious in our assay"

"Of 156 synonymous variants and six missense variants previously reported to be functionally neutral in published literature and/or reported in ClinVar as benign or likely benign (benchmark benign variants), all were characterized as functionally neutral in our assay "

"Of 31 VUSs previously reported to be functionally deleterious, 28 (90.3%) were functionally deleterious and 3 (9.7%) were of indeterminate function in our assay."

"Similarly, of 18 VUSs previously reported to be functionally neutral, 16 (88.9%) were functionally neutral and 2 (11.1%) were of indeterminate function in our assay"

Here we get to the key issues. Variants are generally classified as benign, pathogenic/deleterious, or "variant of unknown/uncertain significance". The latter are particularly vexing to clinical geneticists. The whole point of sequencing a patient's tumor or genomic DNA is to find causal variants that can illuminate their condition, and possibly direct treatment. Seeing lots of "VUS" in the report leaves everyone in the dark. The authors pulled in all the common prediction programs that are officially sanctioned by the ACMG- Americal College of Medical Genetics, which is the foremost guide to clinical genetics, including the functional prediction of otherwise uncharacterized sequence variants. There are seven such programs, including one driven by AI, AlphaMissense that is related to the Nobel prize-winning AlphaFold. 

These programs strain to classify uncharacterized mutations as "likely pathogenic", "likely benign", or, if unable to make a conclusion, VUS/indeterminate. They rely on many kinds of data, like amino acid similarity, protein structure, evolutionary conservation, and known effects in proteins of related structure. They can be extensively validated against known mutations, and against new experimental work as it comes out, so we have a pretty good idea of how they perform. Thus they are trusted to some extent to provide clinical judgements, in the absence of better data. 

Each of seven programs (on bottom) gives estimations of variant effect over the same pool of mutations generated in this paper. This was a weird way to present simple data, but each bar contains the functional results the authors developed in their own data (numbers at the bottom, in parentheses, vertical). The bars were then colored with the rate of deleterious (black) vs benign (white) prediction from the program. The ideal case would be total black for the first bar in each set of three (deleterious) and total white in the third bar in each set (benign). The overall lineup/accuracy of all program predictions vs the author data was then overlaid by a red bar (right axis). The PrimateAI program was specially derived from comparison of homologous genes from primates only, yielding a high-quality dataset about the importance of each coded amino acid. However, it only gave estimates for 906 out of the whole set of 2964 variants. On the other hand, cruder programs like PolyPhen-2 gave less than 40% accuracy, which is quite disappointing for clinical use.

As shown above, the algorithms gave highly variable results, from under 40% accurate to over 80%. It is pretty clear that some of the lesser programs should be phased out. Of programs that fielded all the variants, the best were AlphaMissense and VEST, which each achieved about 70% accuracy. This is still not great. The issue is that, if a whole genome sequence is run for a patient with an obscure disease or syndrome, and variants vs the reference sequence are seen in several hundred genes, then a gene like CDKN2A could easily be pulled into the list of pathogenic (and possibly causal) variants, or be left out, on very shaky evidence. That is why even small increments in accuracy are critically important in this field. Genetic testing is a classic needle-in-a-haystack problem- a quest to find the one mutation (out of millions) that is driving a patient's cancer, or a child's inherited syndrome.

Still outstanding is the issue of non-coding variants. Genes are not just affected by mutations in their protein coding regions (indeed many important genes do not code for proteins at all), but by regulatory regions nearby and far. This is a huge area of mutation effects that are not really algorithmically accessible yet. As a prediction problem, it is far more difficult than predicting effects on a coded protein. It will requiring modeling of the entire gene expression apparatus, much of which remains shrouded in mystery.

• Oh, no! Not the shopping cart!
• Jevons and Malthus will have their say, if we do not get our act together.
• It is time to scrap GDP, and the mania for growth.
• Blaming the victim.
• Everything crypto will be just perfect.
• Can I play piano with my mind?
• We are killers.

The Submission Drive

Humans have a drive for social and intellectual submission, which is extremely dangerous.

There was a time when psychological "drives" were all the rage. The idea that humans have instincts much as other animals do was just entering the scientific consciousness, so finding and classifying them was an important task- the great work of figuring out the human unconscious, or subconscious. Drives for food, security, sex, dominance, and much else were found. Freud even elaborated a "death drive". But our current political epoch suggests another one- a submission drive.

To an independent minded scholar and skeptic, the behavior on the Republican side of the political spectrum is revolting. Falling all over themselves to fawn over a narcissistic megalomaniac? Check. Thinking nothing of flagrant corruption that makes Warren Harding look like a choir boy? Check. Explaining away gross incompetence and pointless cruelty across the entire policy space from economics to foreign policy? Check. What causes people to join and defend what amount to cults? For that matter, what causes people to join religions?

At one level, submission is eminently rational. Groups are always more powerful than individuals. The American archetype of the loner, the Clint Eastwood or John Wayne character riding alone to mete out justice and bucking the system- that is a fantasy. It is powerful precisely because it is so romantic and unrealistic. It is compensatory psychic food for the hemmed-in and submissive. In reality, the system always wins. Militaries win when they can gather up a bigger army than the other guys. Corporations win when they have bought all their competition and become the biggest on the block. Our social instincts lead us to join groups to gain power. 

But the submission drive seems to go way beyond this, allowing us to swallow alternate realities and even seek domination by others. An interesting form is when whole cultures convert their religion. Many times, such as during the colonial era, during the Christian conquests of Northern Europe, and during the Muslim conquests, the winning power foists its religion on another culture, a culture that grows quite rapidly to accept and adopt it as its own. Was one religion true-er or better than the other? Not at all. The new one is often significantly worse in many dimensions than the old. This is purely a power transaction where those who had submitted themselves to one archetype and narrative of cultural and supernatural power find themselves convinced that social and military coercion is a pretty important form of power too, perhaps signifying a new narrative that they should submit to. But once converted, the same psychic events happen. Leaders are idolized, scriptures are memorized, vestments are accessorized. In return, those who submit seek safety and guidance, buying into a (new) father figure archetype.

Joining a group inserts you into a hierarchy of domination. There are rewards for working your way up the ranks, being able to get others to serve you, having more influence and status. This most obvious in the military, with its obsession with colorful gradations, decorations, and uniforms. But it is true everywhere- in corporations, politics, organized crime, families. Submission is the price of entry, and it seems that to properly submit, one has to take on the a great deal more than just a signed contract. Members of organizations are constantly being tested for their loyalty, their buy-in to the ethics and goals of the organization, and its wider world-view. At IBM, they used to sing the company song. Modern corporate life is a complex compromise where some of the submittee's personal life is allowed to be separated from corporate control, and many boundaries are set by legal regime to prevent the organization from turning into a criminal entity and bar total domination of its employees, customers, business partners. 

However, other organizations are not so limited. Religion and politics are a bit less hemmed-in, and demand sometimes extraordinary kinds of fealty for the rewards on offer. In their variety of styles and cultures, they attract different temperaments of devotee. Overall, one has to say that people more prone to submission and participation in hierarchies tend to go to right-wing political, military, and religious organizations. Contrary to the cultivated image of hard-headedness and independence, conservatives turn out to submit more readily to domination by others. It is notorious that organizing Democrats is like herding cats. Likewise, university faculty tend towards independence and disorganization. Liberal churches are notoriously light on discipline and free with their theology. 

Conversely, Republican and conservative organizations spring up like weeds and have, aside from gobs of funding, remarkable discipline. The MAGA swoon for the current president is just one example of the lengths to which thought patterns can be bent in favor of the dominant leader of the moment. The corollary of greater mental submission by the followers is greater rewards and wider scope of action for the leaders. Making it to the top of such disciplined heap seems to turn psychology on its head, from submission to domination. Napoleon is a case study, working his way up the ranks, literally, to a position of ultimate power. Which promptly went to his head, causing him to veer in a conservative direction, and to wreck half of Europe. Cult leaders have time and again shown how poorly adapted we are to this much-sought after, but rarely successful, psychological transition.

The fascist/authoritarian moment that is glowering around the world has reactivated these extreme domination/submission dynamics, such as between Russia and Ukraine, and within so many far-right movements and the poitical systems they target. Fortunately, there are just fundamental temperamental barriers to the attractiveness of such movements, forcing them to take extra-legal measures if they are truly dedicated to overcome the resistence of the less submissive members of their societies.

• Much of this is drawn from a classic book, "Instincts of the Herd in Peace and War".
• Report card on the courts.
• Waymo is hitting the highway.
• Plastic.

Links Only

Due to the press of other activities, only links this week.

• Wingnut medicine.
• Video of a fault slipping during an earthquake... look in the background.
• The original unitary executive.
• How about some free mid-day electricity?
• More Supreme injustice.
• Jay Bhattacharya is a stooge.

Modeling Human Attention

How attention works in the brain is becoming clearer through empirical and computer modeling work.

The current World Series is a tour de force of mental concentration and attention. Batters intently watch a pitch, and have milliseconds to decide that it isn't any good. Pitchers study the opposing batters for any signs of gullibility. Managers face excruciating decisions on when to pull a pitcher in danger. Spectators decide whether to get drawn into the pitcher's duel, or chat with their neighbors. Advertisers measure attention in dollars and cents.

The economics of attention may have reached a fever pitch, but the physiology of attention is only slowly being revealed. Attention is obviously closely aligned with consciousness, so progress on one implies progress on the other as well. The current paper is a computer modeling project, trying to simulate the core connections and behavior between the thalamus and cortex that are involved in sensory perception. For instance, mice are given a slight push on a whisker. If they respond to that, it shows they perceived it. At a very light threshold level, the chances of perception can tuned to 50%, and perceptual events can have more to do with the mental status of the mouse and the history of whisker stimulation than it does with the (consistent) strength of the stimulus. A similar threshold phenomenon holds in other forms of sensing and perception, such as binocular rivalry. Indeed, in binocular rivalry of vision, there is a slow switching back and forth between each image, based on neural accommodation after a few seconds attending to one of the images. Such threshold levels of perception are the bread and butter of research on attention.

"Given the ubiquity of the thalamocortical circuit architecture across sensory modalities, we, along with others, have proposed that reverberant bursting activity in L5PT [thick-tufted layer 5 pyramidal-tract neurons]– matrix thalamus loops may be a necessary component part in a domain general mechanism of perceptual awareness."

"Optogenetic excitation of the apical dendrites reduced the animal’s threshold for awareness, increasing both hits and false-alarms. In turn, pharmocological inhibition of the apical dendrites and POm [posteromedial thalamic nucleus] (a matrix-rich higher-order thalamic nucleus with closed loop connections to barrel cortex) increased the animal's perceptual threshold."

The model was based on the physiology of thalamo-cortical loops, which are very common in sensory, motor, and other circuits. Attention is not something that "happens" in one place, but rather appears to be a state of the network, after negotiation between upper and lower levels. Strong stimuli, such as the roar of the crowd after a home run, push their way to the top of the attention chain. Conversely, a quest for a hot dog can lead to highly focused top-down attention on planning and making a trip to the concession stand, while ignoring everything else going on in the game.

A bit of physiology, showing how mouse neurons connect between cortical and thalamus levels, and how they look at various levels within the cortical sheet. 

These researchers found that by making a faithful model of the neurons and connections found by physiology, they could then functionally model quite faithfully the actions of this loop, including its perceptual thresholds, stochastic activation, and tendency to accommodate (dampen) repeated stimuli. A key aspect of the physiology is the layering of the cortex. Evolution has left clear marks in brain areas of various vintages, in the form of layer organization. The most primitive areas like the brain stem and cerebellum have no layering at all, but rather have anatomical structures, bulbs, and sub-nuclei. Less primitive subcortical (limbic) areas have roughly three layers of cells, whereas the neocortex has a structure that is a uniform sheet, modularized not so much into structural bulbs, knobs, etc, but in a more regularized arrangement of six distinct layers, with columns of activity for parcellation of function. This regular arrangement is replicated all over the cortex and (re-)used for innumerable functions, from sensation and motor control to decision making and emotional control. In general, the middle layer (4) contains the most cell bodies, (called the granular layer), and connects extensively to the other layers. The upper layers (numbered 1,2,3, and closest to the outside of the brain) receive inputs from the thalamus and other areas, while the lower layers (5 and 6 ) send outputs to lower areas of the brain, for motor control, attention control, etc.

The authors simulate cells and connections from the known physiology. Cortical levels shown on left, and an example of one set of connected, active, firing cells on right.

The thalamocortical loop, therefore, as illustrated by the authors, is a cycle of connections from the thalamus into cortical layer 1, which connects to cells stationed in layer 5, which then send axons out back to the thalamus. The thalamus is a large structure nested within/under the hippocampus, basal ganglia, and corpus collosum, that mediates cortical signals to and from the rest of the nervous system, including senses. The authors basically find that they can reproduce the signature bursting behavior of this circuit that others have argued is a sign of attention. That is, the brief set of six pulses between 0.1 and 0.2 seconds above, which is quickly shut down by continuing activity from the inhibitory basket cells (orange, BC). I can't speak to the details either of the modeling or the physiology it is based on, but the authors try their best to hew to realistic cells and physiological circuits, making the case that the neural behavior they get out is a realistic simulation, which can then be used for other perturbations and studies of this system. 

A question that arises is the relationship of consciousness with attention. Are the neural patterns characteristic of attention all that is required to also be conscious? A review of the field says no, they are different, or at least that consciousness is a broader concept that contains attention, but also can contain bare awareness without specific focus. I am not so sure, since attention can presumably be directed inside to our own thoughts and memories, that constitutes the floating kind of awareness of bare consciousness. Without attention to anything, we lack conscious content, and thus, perhaps consciousness itself. The idea that we can meditate our way to a content-less consciousness is time and again disproven by the practice of meditation. It finds that a focus is essential, not to empty the mind of all contents, which I believe is impossible, but to control those mental contents and attain a controlled level of lucid dreaming or mantra-driven reduced consciousness that seems to be the goal of meditation.

The authors of this paper mention that their view of attention is very compatible with each of several reigning theories of consciousness, inchoate as those are, so we do seem to be heading, ever so slowly, towards a solution of this long-standing problem.

• The last time the President and military were at such loggerheads.
• China knows what it is doing.
• Make more market housing.
• The NewsHour asks scientists how they are doing.
• We are going there.